The rapid advancement of video generation models has made it increasingly challenging to distinguish AI-generated videos from real ones. This issue underscores the urgent need for effective AI-generated video detectors to prevent the dissemination of false information via such videos. However, the development of high-performance AI-generated video detectors is currently impeded by the lack of large-scale, high-quality datasets specifically designed for generative video detection. To this end, we introduce GenVidBench, a challenging AI-generated video detection dataset with several key advantages: 1) Large-scale video collection: The dataset contains 6.78 million videos and is currently the largest dataset for AI-generated video detection. 2) Cross-Source and Cross-Generator: The cross-source generation reduces the interference of video content on the detection. The cross-generator ensures diversity in video attributes between the training and test sets, preventing them from being overly similar. 3) State-of-the-Art Video Generators: The dataset includes videos from 11 state-of-the-art AI video generators, ensuring that it covers the latest advancements in the field of video generation. These generators ensure that the datasets are not only large in scale but also diverse, aiding in the development of generalized and effective detection models. Additionally, we present extensive experimental results with advanced video classification models. With GenVidBench, researchers can efficiently develop and evaluate AI-generated video detection models.

Xiaoyu Liu, Yun Zhang, Wei Li, Simiao Li, Xudong Huang, Hanting Chen, Yehui Tang, Jie Hu, Zhiwei Xiong, Yunhe Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

The spectacular success of training large models on extensive datasets highlights the potential of scaling up for exceptional performance. To deploy these models on edge devices, knowledge distillation (KD) is commonly used to create a compact model from a larger, pretrained teacher model. However, as models and datasets rapidly scale up in practical applications, it is crucial to consider the applicability of existing KD approaches originally designed for limited-capacity architectures and small-scale datasets. In this paper, we revisit current KD methods and identify the presence of a small-data pitfall, where most modifications to vanilla KD prove ineffective on large-scale datasets. To guide the design of consistently effective KD methods across different data scales, we conduct a meticulous evaluation of the knowledge transfer process. Our findings reveal that incorporating more useful information is crucial for achieving consistently effective KD methods, while modifications in loss functions show relatively less significance. In light of this, we present a paradigmatic example that combines vanilla KD with deep supervision, incorporating additional information into the student during distillation. This approach surpasses almost all recent KD methods. We believe our study will offer valuable insights to guide the community in navigating beyond the small-data pitfall and toward consistently effective KD.

Multimodal vision language models (VLMs) have made significant progress with the support of continuously increasing model sizes and data volumes. Running VLMs on edge devices has become a challenge for their widespread application. There are several efficient VLM efforts, but they often sacrifice linguistic capabilities to enhance multimodal abilities, or require extensive training. To address this quandary, we introduce the innovative framework of Efficient Vision Language Models with Elastic Visual Experts (Eve). By strategically incorporating adaptable visual expertise at multiple stages of training, Eve strikes a balance between preserving linguistic abilities and augmenting multimodal capabilities. This balanced approach results in a versatile model with only 1.8B parameters that delivers significant improvements in both multimodal and linguistic tasks. Notably, in configurations below 3B parameters, Eve distinctly outperforms in language benchmarks and achieves state-of-the-art results in VLM Benchmarks. Additionally, its multimodal accuracy outstrips that of the larger 7B LLaVA-1.5 model.

Recently segment anything model (SAM) has shown powerful segmentation capability and has drawn great attention in computer vision fields. Massive following works have developed various applications based on the pre-trained SAM and achieved impressive performance on downstream vision tasks. However, SAM consists of heavy architectures and requires massive computational capacity, which hinders the further application of SAM on computation constrained edge devices. To this end, in this paper we propose a framework to obtain a tiny segment anything model (TinySAM) while maintaining the strong zero-shot performance. We first propose a full-stage knowledge distillation method with hard prompt sampling and hard mask weighting strategy to distill a lightweight student model. We also adapt the post-training quantization to the prompt-based segmentation task and further reduce the computational cost. Moreover, a hierarchical segmenting everything strategy is proposed to accelerate the everything inference by 2× with almost no performance degradation. With all these proposed methods, our TinySAM leads to orders of magnitude computational reduction and pushes the envelope for efficient segment anything task. Extensive experiments on various zero-shot transfer tasks demonstrate the significantly advantageous performance of our TinySAM against counterpart methods.

Transformer and its variants have shown great potential for various vision tasks in recent years, including image classification, object detection and segmentation. Meanwhile, recent studies also reveal that with proper architecture design, convolutional networks (ConvNets) also achieve competitive performance with transformers. However, no prior methods have explored to utilize pure convolution to build a Transformer-style Decoder module, which is essential for Encoder-Decoder architecture like Detection Transformer (DETR). To this end, in this paper we explore whether we could build query-based detection and segmentation framework with ConvNets instead of sophisticated transformer architecture. We propose a novel mechanism dubbed InterConv to perform interaction between object queries and image features via convolutional layers. Equipped with the proposed InterConv, we build Detection ConvNet (DECO), which is composed of a backbone and convolutional encoder-decoder architecture. We compare the proposed DECO against prior detectors on the challenging COCO benchmark. Despite its simplicity, our DECO achieves competitive performance in terms of detection accuracy and running speed. Specifically, with the ResNet-18 and ResNet-50 backbone, our DECO achieves $40.5\%$ and $47.8\%$ AP with $66$ and $34$ FPS, respectively. The proposed method is also evaluated on the segment anything task, demonstrating similar performance and higher efficiency. We hope the proposed method brings another perspective for designing architectures for vision tasks. Codes are available at \url{https://github.com/xinghaochen/DECO} and \url{https://github.com/mindspore-lab/models/tree/master/research/huawei-noah/DECO}.

Post-training quantization (PTQ) has played a pivotal role in compressing large language models (LLMs) at ultra-low costs. Although current PTQ methods have achieved promising results by addressing outliers and employing layer- or block-wise loss optimization techniques, they still suffer from significant performance degradation at ultra-low bits precision. To dissect this issue, we conducted an in-depth analysis of quantization errors specific to LLMs and surprisingly discovered that, unlike traditional sources of quantization errors, the growing number of model parameters, combined with the reduction in quantization bits, intensifies inter-layer and intra-layer dependencies, which severely impact quantization accuracy. This finding highlights a critical challenge in quantizing LLMs. To address this, we propose CBQ, a cross-block reconstruction-based PTQ method for LLMs. CBQ leverages a cross-block dependency to establish long-range dependencies across multiple blocks and integrates an adaptive LoRA-Rounding technique to manage intra-layer dependencies. To further enhance performance, CBQ incorporates a coarse-to-fine pre-processing mechanism for processing weights and activations. Extensive experiments show that CBQ achieves superior low-bit quantization (W4A4, W4A8, W2A16) and outperforms existing state-of-the-art methods across various LLMs and datasets. Notably, CBQ only takes 4.3 hours to quantize a weight-only quantization of a 4-bit LLAMA1-65B model, achieving a commendable trade off between performance and efficiency.

Sampling from diffusion models can be seen as solving the corresponding probability flow ordinary differential equation (ODE). The solving process requires a significant number of function evaluations (NFE), making it time-consuming. Recently, several solver search frameworks have attempted to find better-performing model-specific solvers. However, predicting the impact of intermediate solving strategies on final sample quality remains challenging, rendering the search process inefficient. In this paper, we propose a novel method for designing solving strategies. We first introduce a unified prediction formula for linear multistep solvers. Subsequently, we present a solver distillation framework, which enables a student solver to mimic the sampling trajectory generated by a teacher solver with more steps. We utilize the mean Euclidean distance between the student and teacher sampling trajectories as a metric, facilitating rapid adjustment and optimization of intermediate solving strategies. The design space of our framework encompasses multiple aspects, including prediction coefficients, time step schedules, and time scaling factors. Our framework has the ability to complete a solver search for Stable-Diffusion in under 12 total GPU hours. Compared to previous reinforcement learning-based search frameworks, our approach achieves over a 10$\times$ increase in search efficiency. With just 5 NFE, we achieve FID scores of 3.23 on CIFAR10, 7.16 on ImageNet-64, 5.44 on LSUN-Bedroom, and 12.52 on MS-COCO, resulting in a 2$\times$ sampling acceleration ratio compared to handcrafted solvers.

While large-scale training data is fundamental for developing capable large language models (LLMs), strategically selecting high-quality data has emerged as a critical approach to enhance training efficiency and reduce computational costs. Current data selection methodologies predominantly rely on static, training-agnostic criteria, failing to account for the dynamic model training and data interactions. In this paper, we propose a new Data Weighting Model (DWM) to adjust the weight of selected data within each batch to achieve a dynamic data utilization during LLM training. Specially, to better capture the dynamic data preference of the trained model, a bi-level optimization framework is implemented to update the weighting model. Our experiments demonstrate that DWM enhances the performance of models trained with randomly-selected data, and the learned weighting model can be transferred to enhance other data selection methods and models of different sizes. Moreover, we further analyze how a model’s data preferences evolve throughout training, providing new insights into the data preference of the model during training.

Large Language Models (LLMs) have demonstrated remarkable abilities across various language tasks, but solving complex reasoning problems remains a significant challenge. While existing methods, such as Chain-of-Thought (CoT) and Tree-of-Thought (ToT), enhance reasoning by decomposing problems or structuring prompts, they typically perform a single pass of reasoning and may fail to revisit flawed paths, compromising accuracy. To address this limitation, we propose a novel reasoning framework called Forest-of-Thought (FoT), which integrates multiple reasoning trees to leverage collective decision-making for solving complex logical problems. FoT employs sparse activation strategies to select the most relevant reasoning paths, improving both efficiency and accuracy. Additionally, we introduce a dynamic self-correction strategy that enables real-time error correction, along with consensus-guided decision-making strategies to optimize both correctness and computational resources. Experimental results demonstrate that the FoT framework, combined with these strategies, significantly enhances the reasoning capabilities of LLMs, enabling them to solve complex tasks with greater precision and efficiency.

Large language models(LLMs) have garnered significant attention and demonstrated impressive capabilities in a wide range of applications. However, due to their enormous computational costs, the deployment and application of LLMs are often severely limited. To address this issue, structured pruning is an effective solution to compress the parameters of LLMs. Determining the importance of each sub-module in LLMs and minimizing performance loss are critical issues that need to be carefully addressed in structured pruning. In this paper, we propose an effective and fast structured pruning method named SlimLLM for large language models. For channel and attention head pruning, we evaluate the importance based on the entire channel or head, rather than merely aggregating the importance of individual elements within a sub-module. This approach enables a more holistic consideration of the interdependence among elements within the sub-module. In addition, we design a simple linear regression strategy for the output matrix to quickly recover performance. We also propose layer-based importance ratio to determine the pruning ratio for each layer. Based on the LLaMA benchmark results, our SlimLLM outperforms other methods and achieves state-of-the-art performance.

Mixture-of-Experts (MoE) activates only a subset of experts during inference, allowing the model to maintain low inference FLOPs and latency even as the parameter count scales up. However, since MoE dynamically selects the experts, all the experts need to be loaded into VRAM. Their large parameter size still limits deployment, and offloading, which load experts into VRAM only when needed, significantly increase inference latency. To address this, we propose Mixture of Lookup Experts (MoLE), a new MoE architecture that is efficient in both communication and VRAM usage. In MoLE, the experts are Feed-Forward Networks (FFNs) during training, taking the output of the embedding layer as input. Before inference, these experts can be re-parameterized as lookup tables (LUTs) that retrieves expert outputs based on input ids, and offloaded to storage devices. Therefore, we do not need to perform expert computations during inference. Instead, we directly retrieve the expert’s computation results based on input ids and load them into VRAM, and thus the resulting communication overhead is negligible. Experiments show that, with the same FLOPs and VRAM usage, MoLE achieves inference speeds comparable to dense models and significantly faster than MoE with experts offloading, while maintaining performance on par with MoE. Code: https://github.com/JieShibo/MoLE.

Wei He, Kai Han, Yehui Tang, Chengcheng Wang, Yujie Yang, Tianyu Guo, Yunhe Wang. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Yunsheng Ni, Chuanjian Liu, Yehui Tang, Kai Han, Yunhe Wang. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Ying Nie, Binwei Yan, Tianyu Guo, Hao Liu, Haoyu Wang, Wei He, Binfan Zheng, Weihao Wang, Qiang Li, Weijian Sun, Yunhe Wang, Dacheng Tao. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Representation Alignment (REPA) that aligns Diffusion Transformer (DiT) hidden-states with ViT visual encoders has proven highly effective in DiT training, demonstrating superior convergence properties, but it has not been validated on the canonical diffusion U-Net architecture that shows faster convergence compared to DiTs. However, adapting REPA to U-Net architectures presents unique challenges: (1) different block functionalities necessitate revised alignment strategies; (2) spatial-dimension inconsistencies emerge from U-Net's spatial downsampling operations; (3) space gaps between U-Net and ViT hinder the effectiveness of tokenwise alignment. To encounter these challenges, we propose U-REPA, a representation alignment paradigm that bridges U-Net hidden states and ViT features as follows: Firstly, we propose via observation that due to skip connection, the middle stage of U-Net is the best alignment option. Secondly, we propose upsampling of U-Net features after passing them through MLPs. Thirdly, we observe difficulty when performing tokenwise similarity alignment, and further introduces a manifold loss that regularizes the relative similarity between samples. Experiments indicate that the resulting U-REPA could achieve excellent generation quality and greatly accelerates the convergence speed. With CFG guidance interval, U-REPA could reach FID<1.5 in 200 epochs or 1M iterations on ImageNet 256 $\times$ 256, and needs only half the total epochs to perform better than REPA under \textit{sd-vae-ft-ema}.

Deep Neural Networks (DNNs) are vulnerable to adversarial attacks. Existing methods are devoted to developing various robust training strategies or regularizations to update the weights of the neural network. But beyond the weights, the overall structure and information flow in the network are explicitly determined by the neural architecture, which remains unexplored. Thus, this paper aims to improve the adversarial robustness of the network from the architectural perspective. We explore the relationship among adversarial robustness, Lipschitz constant, and architecture parameters and show that an appropriate constraint on architecture parameters could reduce the Lipschitz constant to further improve the robustness. The importance of architecture parameters could vary from operation to operation or connection to connection. We approximate the Lipschitz constant of the entire network through a univariate log-normal distribution, whose mean and variance are related to architecture parameters. The confidence can be fulfilled through formulating a constraint on the distribution parameters based on the cumulative function. Compared with adversarially trained neural architectures searched by various NAS algorithms as well as efficient human-designed models, our algorithm empirically achieves the best performance among all the models under various attacks on different datasets.

Recent advancements in vision transformers (ViTs) have demonstrated that larger models often achieve superior performance. However, training these models remains computationally intensive and costly. To address this challenge, we introduce ScaleNet, an efficient approach for scaling ViT models. Unlike conventional training from scratch, ScaleNet facilitates rapid model expansion with negligible increases in parameters, building on existing pretrained models. This offers a cost-effective solution for scaling up ViTs. Specifically, ScaleNet achieves model expansion by inserting additional layers into pretrained ViTs, utilizing layer-wise weight sharing to maintain parameters efficiency. Each added layer shares its parameter tensor with a corresponding layer from the pretrained model. To mitigate potential performance degradation due to shared weights, ScaleNet introduces a small set of adjustment parameters for each layer. These adjustment parameters are implemented through parallel adapter modules, ensuring that each instance of the shared parameter tensor remains distinct and optimized for its specific function. Experiments on the ImageNet-1K dataset demonstrate that ScaleNet enables efficient expansion of ViT models. With a $2\times $ depth-scaled DeiT-Base model, ScaleNet achieves a 7.42% accuracy improvement over training from scratch while requiring only one-third of the training epochs, highlighting its efficiency in scaling ViTs. Beyond image classification, our method shows significant potential for application in downstream vision areas, as evidenced by the validation in object detection task.

The encoder-decoder architecture is a prevailing paradigm for semantic segmentation. It has been discovered that aggregation of multi-stage encoder features plays a significant role in capturing discriminative pixel representation. In this work, we rethink feature reconstruction for scale alignment of multi-stage pyramidal features and treat it as a Query Update (Q-UP) task. Pixel-wise affinity scores are calculated between the high-resolution query map and low-resolution feature map to dynamically broadcast low-resolution pixel features to match a higher resolution. Unlike prior works (e.g. bilinear interpolation) that only exploit sub-pixel neighborhoods, Q-UP samples contextual information within a global receptive field via a data-dependent manner. To alleviate intra-category feature variance, we substitute source pixel features for feature reconstruction with their corresponding category prototype that is assessed by averaging all pixel features belonging to that category. Besides, a memory module is proposed to explore the capacity of category prototypes at the dataset level. We refer to the method as Category Prototype Transformer (CPT). We conduct extensive experiments on popular benchmarks. Integrating CPT into a feature pyramid structure exhibits superior performance for semantic segmentation even with low-resolution feature maps, e.g. 1/32 of the input size, significantly reducing computational complexity. Specifically, the proposed method obtains a compelling 55.5% mIoU with greatly reduced model parameters and computations on the challenging ADE20K dataset.

The large-scale visual pretraining has significantly improve the performance of large vision models. However, we observe the low FLOPs pitfall that the existing low-FLOPs models cannot benefit from large-scale pretraining. In this paper, we introduce a novel design principle, termed ParameterNet, aimed at augmenting the number of parameters in large-scale visual pretraining models while minimizing the increase in FLOPs. We leverage dynamic convolutions to incorporate additional parameters into the networks with only a marginal rise in FLOPs. The ParameterNet approach allows low-FLOPs networks to take advantage of large-scale visual pretraining. Furthermore, we extend the ParameterNet concept to the language domain to enhance inference results while preserving inference speed. Experiments on the large-scale ImageNet-22K have shown the superiority of our ParameterNet scheme. For example, ParameterNet-600M can achieve higher accuracy than the widely-used Swin Transformer (81.6% vs. 80.9%) and has much lower FLOPs (0.6G vs. 4.5G). The code will be released at https://parameternet.github.io/.

The laws of model size, data volume, computation and model performance have been extensively studied in the field of Natural Language Processing (NLP). However, the scaling laws in Scene Text Recognition (STR) have not yet been investigated. To address this, we conducted comprehensive studies that involved examining the correlations between performance and the scale of models, data volume and computation in the field of text recognition. Conclusively, the study demonstrates smooth power laws between performance and model size, as well as training data volume, when other influencing factors are held constant. Additionally, we have constructed a large-scale dataset called REBU-Syn, which comprises 6 M real samples and 18 M synthetic samples. Based on the disclosed scaling law and new dataset, we successfully trained a scene text recognition model, achieving a new state-of-the-art on 6 common test benchmarks with top-1 average accuracy of 97.42%. The models and dataset are publicly available at large-ocr-model.github.io.

Super-Resolution (SR) reconstructs high-resolution images from low-resolution ones. CNNs and window-attention methods are two major categories of canonical SR models. However, these measures are rigid: in both operations, each pixel gathers the same number of neighboring pixels, hindering their effectiveness in SR tasks. Alternatively, we leverage the flexibility of graphs and propose the Image Processing GNN (IPG) model to break the rigidity that dominates previous SR methods. Firstly, SR is unbalanced in that most reconstruction efforts are concentrated to a small proportion of detail-rich image parts. Hence, we leverage degree flexibility by assigning higher node degrees to detail-rich image nodes. Then in order to construct graphs for SR-effective aggregation, we treat images as pixel node sets rather than patch nodes. Lastly, we hold that both local and global information are crucial for SR performance. In the hope of gathering pixel information from both local and global scales efficiently via flexible graphs, we search node connections within nearby regions to construct local graphs; and find connections within a strided sampling space of the whole image for global graphs. The flexibility of graphs boosts the SR performance of the IPG model. Experiment results on various datasets demonstrates that the proposed IPG outperforms State-of-the-Art baselines. Codes are available at this link.

In image restoration (IR), leveraging semantic priors from segmentation models has been a common approach to improve performance. The recent segment anything model (SAM) has emerged as a powerful tool for extracting advanced semantic priors to enhance IR tasks. However, the computational cost of SAM is prohibitive for IR, compared to existing smaller IR models. The incorporation of SAMfor extracting semantic priors considerably hampers the model inference efficiency. To address this issue, we propose a general framework to distill SAM's semantic knowledge to boost exiting IR models without interfering with their inference process. Specifically, our proposed framework consists of the semantic priors fusion (SPF) scheme and the semantic priors distillation (SPD) scheme. SPF fuses two kinds of information between the restored image predicted by the original IR model and the semantic mask predicted by SAM for the refined restored image. SPD leverages a self-distillation manner to distill the fused semantic priors to boost the performance of original IR models. Additionally, we design a semantic-guided relation (SGR) module for SPD, which ensures semantic feature representation space consistency to fully distill the priors. We demonstrate the effectiveness of our framework across multiple IR models and tasks, including deraining, deblurring, and denoising.

Existing text-based person retrieval datasets often have relatively coarse-grained text annotations. This hinders the model to comprehend the fine-grained semantics of query texts in real scenarios. To address this problem, we con-tribute a new benchmark named UFineBench for text-based person retrieval with ultra-fine granularity. Firstly, we construct a new dataset named UFine6926. We collect a large number of person images and manually annotate each image with two detailed textual descriptions, averaging 80.8 words each. The average word count is three to four times that of the previous datasets. In addition of standard in-domain evaluation, we also propose a spe-cial evaluation paradigm more representative of real sce-narios. It contains a new evaluation set with cross domains, cross textual granularity and cross textual styles, named UFine3C, and a new evaluation metric for accurately mea-suring retrieval ability, named mean Similarity Distribution (mSD). Moreover, we propose CFAM, a more efficient al-gorithm especially designed for text-based person retrieval with ultra fine-grained texts. It achieves fine granularity mining by adopting a shared cross-modal granularity de-coder and hard negative match mechanism. With standard in-domain evaluation, CFAM establishes competitive performance across various datasets, espe-cially on our ultra fine-grained UFine6926. Furthermore, by evaluating on UFine3C, we demonstrate that training on our UFine6926 significantly improves generalization to real scenarios compared with other coarse-grained datasets. The dataset and code will be made publicly available at https://github.com/Zplusdragon/UFineBench.

With the advancement of generative modeling techniques, synthetic human speech becomes increasingly indistinguishable from real, and tricky challenges are elicited for the audio deepfake detection (ADD) system. In this paper, we exploit audio features to improve the generalizability of ADD systems. Investigation of the ADD task performance is conducted over a broad range of audio features, including various handcrafted features and learning-based features. Experiments show that learning-based audio features pretrained on a large amount of data generalize better than hand-crafted features on out-of-domain scenarios. Subsequently, we further improve the generalizability of the ADD system using proposed multi-feature approaches to incorporate complimentary information from features of different views. The model trained on ASV2019 data achieves an equal error rate of 24.27% on the In-the-Wild dataset. The code will be released as soon1.

Recent releases of Large Language Models (LLMs), e.g. ChatGPT, are astonishing at generating human-like texts, but they may impact the authenticity of texts. Previous works proposed methods to detect these AI-generated texts, including simple ML classifiers, pretrained-model-based zero-shot methods, and finetuned language classification models. However, mainstream detectors always fail on short texts, like SMSes, Tweets, and reviews. In this paper, a Multiscale Positive-Unlabeled (MPU) training framework is proposed to address the difficulty of short-text detection without sacrificing long-texts. Firstly, we acknowledge the human-resemblance property of short machine texts, and rephrase AI text detection as a partial Positive-Unlabeled (PU) problem by regarding these short machine texts as partially "unlabeled". Then in this PU context, we propose the length-sensitive Multiscale PU Loss, where a recurrent model in abstraction is used to estimate positive priors of scale-variant corpora. Additionally, we introduce a Text Multiscaling module to enrich training corpora. Experiments show that our MPU method augments detection performance on long AI-generated texts, and significantly improves short-text detection of language model detectors. Language Models trained with MPU could outcompete existing detectors on various short-text and long-text detection benchmarks. The codes are available at https://github.com/mindspore-lab/mindone/tree/master/examples/detect_chatgpt and https://github.com/YuchuanTian/AIGC_text_detector.

In an effort to reduce the computational load of Transformers, research on linear attention has gained significant momentum. However, the improvement strategies for attention mechanisms typically necessitate extensive retraining, which is impractical for large language models with a vast array of parameters. In this paper, we present DiJiang, a novel Frequency Domain Kernelization approach that enables the transformation of a pre-trained vanilla Transformer into a linear complexity model with little training costs. By employing a weighted Quasi-Monte Carlo method for sampling, the proposed approach theoretically offers superior approximation efficiency. To further reduce the training computational complexity, our kernelization is based on Discrete Cosine Transform (DCT) operations. Extensive experiments demonstrate that the proposed method achieves comparable performance to the original Transformer, but with significantly reduced training costs and much faster inference speeds. Our DiJiang-7B achieves comparable performance with LLaMA2-7B on various benchmark while requires only about 1/50 training cost. Code is available at https://github.com/YuchuanTian/DiJiang.

Transformers have become foundational architectures for both natural language and computer vision tasks. However, the high computational cost makes it quite challenging to deploy on resource-constraint devices. This paper investigates the computational bottleneck modules of efficient transformer, i.e., normalization layers and attention modules. LayerNorm is commonly used in transformer architectures but is not computational friendly due to statistic calculation during inference. However, replacing LayerNorm with more efficient BatchNorm in transformer often leads to inferior performance and collapse in training. To address this problem, we propose a novel method named PRepBN to progressively replace LayerNorm with re-parameterized BatchNorm in training. Moreover, we propose a simplified linear attention (SLA) module that is simple yet effective to achieve strong performance. Extensive experiments on image classification as well as object detection demonstrate the effectiveness of our proposed method. For example, our SLAB-Swin obtains $83.6%$ top-1 accuracy on ImageNet-1K with $16.2$ms latency, which is $2.4$ms less than that of Flatten-Swin with $0.1%$ higher accuracy. We also evaluated our method for language modeling task and obtain comparable performance and lower latency. Codes are publicly available at https://github.com/xinghaochen/SLAB and https://github.com/mindspore-lab/models/tree/master/research/huawei-noah/SLAB.

Current solutions for efficiently constructing large vision-language (VL) models follow a two-step paradigm: projecting the output of pre-trained vision encoders to the input space of pre-trained language models as visual prompts; and then transferring the models to downstream VL tasks via end-to-end parameter-efficient fine-tuning (PEFT). However, this paradigm still exhibits inefficiency since it significantly increases the input length of the language models. In this paper, in contrast to integrating visual prompts into inputs, we regard visual prompts as additional knowledge that facilitates language models in addressing tasks associated with visual information. Motivated by the finding that Feed-Forward Network (FFN) of language models acts as "key-value memory", we introduce a novel approach termed memory-space visual prompting (MemVP), wherein visual prompts are concatenated with the weights of FFN for visual knowledge injection. Experimental results across various VL tasks and language models reveal that MemVP significantly reduces the training time and inference latency of the finetuned VL models and surpasses the performance of previous PEFT methods.

Large language models (LLM) have recently attracted significant attention in the field of artificial intelligence. However, the training process of these models poses significant challenges in terms of computational and storage capacities, thus compressing checkpoints has become an urgent problem. In this paper, we propose a novel Extreme Checkpoint Compression (ExCP) framework, which significantly reduces the required storage of training checkpoints while achieving nearly lossless performance. We first calculate the residuals of adjacent checkpoints to obtain the essential but sparse information for higher compression ratio. To further excavate the redundancy parameters in checkpoints, we then propose a weight-momentum joint shrinking method to utilize another important information during the model optimization, i.e., momentum. In particular, we exploit the information of both model and optimizer to discard as many parameters as possible while preserving critical information to ensure optimal performance. Furthermore, we utilize non-uniform quantization to further compress the storage of checkpoints. We extensively evaluate our proposed ExCP framework on several models ranging from 410M to 7B parameters and demonstrate significant storage reduction while maintaining strong performance. For instance, we achieve approximately $70\times$ compression for the Pythia-410M model, with the final performance being as accurate as the original model on various downstream tasks. Codes will be available at https://github.com/Gaffey/ExCP.

The power of large language models (LLMs) has been demonstrated through numerous data and computing resources. However, the application of language models on mobile devices is facing huge challenge on the computation and memory costs, that is, tiny language models with high performance are urgently required. Limited by the highly complex training process, there are many details for optimizing language models that are seldom studied carefully. In this study, based on a tiny language model with 1B parameters, we carefully design a series of empirical study to analyze the effect of each component. Three perspectives are mainly discussed, i.e., neural architecture, parameter initialization, and optimization strategy. Several design formulas are empirically proved especially effective for tiny language models, including tokenizer compression, architecture tweaking, parameter inheritance and multiple-round training. Then we train PanGu-$\pi$-1B Pro and PanGu-$\pi$-1.5B Pro on 1.6T multilingual corpora, following the established formulas. Experimental results demonstrate the improved optimization and architecture yield a notable average improvement of 8.87 on benchmark evaluation sets for PanGu-$\pi$-1B Pro. Besides, PanGu-$\pi$-1.5B Pro surpasses a range of SOTA models with larger model sizes, validating its superior performance. The code will be released soon. The code is available at https://github.com/YuchuanTian/RethinkTinyLM.

In order to reduce the computational complexity of large language models, great efforts have been made to to improve the efficiency of transformer models such as linear attention and flash-attention. However, the model size and corresponding computational complexity are constantly scaled up in pursuit of higher performance. In this work, we present MemoryFormer, a novel transformer architecture which significantly reduces the computational complexity (FLOPs) from a new perspective. We eliminate nearly all the computations of the transformer model except for the necessary computation required by the multi-head attention operation. This is made possible by utilizing an alternative method for feature transformation to replace the linear projection of fully-connected layers. Specifically, we first construct a group of in-memory lookup tables that store a large amount of discrete vectors to replace the weight matrix used in linear projection. We then use a hash algorithm to retrieve a correlated subset of vectors dynamically based on the input embedding. The retrieved vectors combined together will form the output embedding, which provides an estimation of the result of matrix multiplication operation in a fully-connected layer. Compared to conducting matrix multiplication, retrieving data blocks from memory is a much cheaper operation which requires little computations. We train MemoryFormer from scratch and conduct extensive experiments on various benchmarks to demonstrate the effectiveness of the proposed model.

Speculative decoding has demonstrated its effectiveness in accelerating the inference of large language models (LLMs) while maintaining an identical sampling distribution. However, the conventional approach of training separate draft model to achieve a satisfactory token acceptance rate can be costly and impractical. In this paper, we propose a novel self-speculative decoding framework \emph{Kangaroo} with \emph{double} early exiting strategy, which leverages the shallow sub-network and the \texttt{LM Head} of the well-trained target LLM to construct a self-drafting model. Then, the self-verification stage only requires computing the remaining layers over the \emph{early-exited} hidden states in parallel. To bridge the representation gap between the sub-network and the full model, we train a lightweight and efficient adapter module on top of the sub-network. One significant challenge that comes with the proposed method is that the inference latency of the self-draft model may no longer be negligible compared to the big model. To boost the token acceptance rate while minimizing the latency of the self-drafting model, we introduce an additional \emph{early exiting} mechanism for both single-sequence and the tree decoding scenarios. Specifically, we dynamically halt the small model's subsequent prediction during the drafting phase once the confidence level for the current step falls below a certain threshold. This approach reduces unnecessary computations and improves overall efficiency. Extensive experiments on multiple benchmarks demonstrate our effectiveness, where Kangaroo achieves walltime speedups up to 2.04$\times$, outperforming Medusa-1 with 88.7\% fewer additional parameters. The code for Kangaroo is available at https://github.com/Equationliu/Kangaroo.

Diffusion Transformers (DiTs) introduce the transformer architecture to diffusion tasks for latent-space image generation. With an isotropic architecture that chains a series of transformer blocks, DiTs demonstrate competitive performance and good scalability; but meanwhile, the abandonment of U-Net by DiTs and their following improvements is worth rethinking. To this end, we conduct a simple toy experiment by comparing a U-Net architectured DiT with an isotropic one. It turns out that the U-Net architecture only gain a slight advantage amid the U-Net inductive bias, indicating potential redundancies within the U-Net-style DiT. Inspired by the discovery that U-Net backbone features are low-frequency-dominated, we perform token downsampling on the query-key-value tuple for self-attention and bring further improvements despite a considerable amount of reduction in computation. Based on self-attention with downsampled tokens, we propose a series of U-shaped DiTs (U-DiTs) in the paper and conduct extensive experiments to demonstrate the extraordinary performance of U-DiT models. The proposed U-DiT could outperform DiT-XL with only 1/6 of its computation cost. Codes are available at https://github.com/YuchuanTian/U-DiT.

Tokenizers serve as crucial interfaces between models and linguistic data, substantially influencing the efficacy and precision of large language models (LLMs). Traditional tokenization methods often rely on static frequency-based statistics and are not inherently synchronized with LLM architectures, which may limit model performance. In this study, we propose a simple but effective method to learn tokenizers specifically engineered for seamless integration with LLMs. Initiating with a broad initial vocabulary, we refine our tokenizer by monitoring changes in the model’s perplexity during training, allowing for the selection of a tokenizer that is closely aligned with the model’s evolving dynamics. Through iterative refinement, we develop an optimized tokenizer. Our empirical evaluations demonstrate that this adaptive approach significantly enhances accuracy compared to conventional methods, maintaining comparable vocabulary sizes and affirming its potential to improve LLM functionality.

The efficacy of large language models (LLMs) on downstream tasks usually hinges on instruction tuning, which relies critically on the quality of training data. Unfortunately, collecting high-quality and diverse data is both expensive and time-consuming. To mitigate this issue, we propose a novel Star-Agents framework, which automates the enhancement of data quality across datasets through multi-agent collaboration and assessment. The framework adopts a three-pronged strategy. It initially generates diverse instruction data with multiple LLM agents through a bespoke sampling method. Subsequently, the generated data undergo a rigorous evaluation using a dual-model method that assesses both difficulty and quality. Finaly, the above process evolves in a dynamic refinement phase, where more effective LLMs are prioritized, enhancing the overall data quality. Our empirical studies, including instruction tuning experiments with models such as Pythia and LLaMA, demonstrate the effectiveness of the proposed framework. Optimized datasets have achieved substantial improvements, with an average increase of 12\% and notable gains in specific metrics, such as a 40\% improvement in Fermi, as evidenced by benchmarks like MT-bench, Vicuna bench, and WizardLM testset. Codes will be released soon.

The success of modern single image super-resolution (SISR) algorithms is inspired by the development of deep convolutional neural networks (CNNs). However, these CNN-based methods require considerable computation and complexity, making it impossible for these methods to perform real-time calculations in edge devices. Thus, lightweight model design has become a development trend in the super-resolution field, including pruning, quantization, and other methods. The 1-bit quantization is an extreme lightweight method which can reduce the calculation amount of the model in an extreme manner and is friendly to hardware such as edge devices. Most existing binary quantization approaches lead to a large information loss during forward propagation, especially in detailed color information (e.g., edge, texture, and contrast). The loss of color information makes modern binary methods unsuitable for SISR tasks. We think the loss occurs because these methods typically utilize a uniform threshold to quantize the weights and activations. Thus, in this article, we thoroughly analyze the difference between normal classification tasks and SISR tasks, and present a binarization scheme based on local means. The proposed method can maintain more detailed information in feature maps using dynamic thresholds during quantization. Specifically, each value in the full precision activations has a corresponding threshold during the quantization process, and those thresholds are determined by the full precision values of the surroundings. In addition, a gradient approximator is introduced to adaptively optimize the gradient for updating binary weights. We then verify the effectiveness of our method for training binary networks on several SISR benchmarks including VDSR and SRResNet. Experimental results show that the proposed method can outperform the state-of-the-art algorithms to obtain binary networks for image super-resolution with better peak signal-to-noise ratio (PSNR) values and visual quality.

Recently, the sizes of deep neural networks and training datasets both increase drastically to pursue better performance in a practical sense. With the prevalence of transformer-based models in vision tasks, even more pressure is laid on the GPU platforms to train these heavy models, which consumes a large amount of time and computing resources as well. Therefore, it's crucial to accelerate the training process of deep neural networks. In this paper, we propose a general network expansion method to reduce the practical time cost of the model training process. Specifically, we utilize both width- and depth-level sparsity of dense models to accelerate the training of deep neural networks. Firstly, we pick a sparse sub-network from the original dense model by reducing the number of parameters as the starting point of training. Then the sparse architecture will gradually expand during the training procedure and finally grow into a dense one. We design different expanding strategies to grow CNNs and ViTs respectively, due to the great heterogeneity in between the two architectures. Our method can be easily integrated into popular deep learning frameworks, which saves considerable training time and hardware resources. Extensive experiments show that our acceleration method can significantly speed up the training process of modern vision models on general GPU devices with negligible performance drop (e.g. 1.42× faster for ResNet-101 and 1.34× faster for DeiT-base on ImageNet-1k). The code is available at https://github.com/huawei-noah/Efficient-Computing/tree/master/TrainingAcceleration/NetworkExpansion and https://gitee.com/mindspore/hub/blob/master/mshub_res/assets/noah-cvlab/gpu/1.8/networkexpansion_v1.0_imagenet2012.md

We present Reference-guided Super-Resolution Neural Radiance Field (RefSR-NeRF) that extends NeRF to super resolution and photorealistic novel view synthesis. Despite NeRF's extraordinary success in the neural rendering field, it suffers from blur in high resolution rendering because its inherent multilayer perceptron struggles to learn high frequency details and incurs a computational explosion as resolution increases. Therefore, we propose RefSR-NeRF, an end-to-end framework that first learns a low resolution NeRF representation, and then reconstructs the high frequency details with the help of a high resolution reference image. We observe that simply introducing the pre-trained models from the literature tends to produce unsatisfied artifacts due to the divergence in the degradation model. To this end, we design a novel lightweight RefSR model to learn the inverse degradation process from NeRF renderings to target HR ones. Extensive experiments on multiple benchmarks demonstrate that our method exhibits an impressive trade-off among rendering quality, speed, and memory usage, outperforming or on par with NeRF and its variants while being$52\times$speedup with minor extra memory usage. Code will be available at: Mindspore and Pytorch

Model quantization is a crucial step for deploying super resolution (SR) networks on mobile devices. However, existing works focus on quantization-aware training, which requires complete dataset and expensive computational overhead. In this paper, we study post-training quantization (PTQ) for image super resolution using only a few unlabeled calibration images. As the SR model aims to maintain the texture and color information of input images, the distribution of activations are long-tailed, asymmetric and highly dynamic compared with classification models. To this end, we introduce the density-based dual clipping to cut off the outliers based on analyzing the asymmetric bounds of activations. Moreover, we present a novel pixel aware calibration method with the supervision of the full-precision model to accommodate the highly dynamic range of different samples. Extensive experiments demonstrate that the proposed method significantly outperforms existing PTQ algorithms on various models and datasets. For instance, we get a 2.091 dB increase on Urban100 benchmark when quantizing$EDSR\times4$to 4-bit with 100 unlabeled images. Our code is available at both PyTorch and MindSpore.

Masked Image Modeling (MIM) achieves outstanding success in self-supervised representation learning. Unfortunately, MIM models typically have huge computational burden and slow learning process, which is an inevitable obstacle for their industrial applications. Although the lower layers play the key role in MIM, existing MIM models conduct reconstruction task only at the top layer of encoder. The lower layers are not explicitly guided and the interaction among their patches is only used for calculating new activations. Considering the reconstruction task requires non-trivial inter-patch interactions to reason target signals, we apply it to multiple local layers including lower and upper layers. Further, since the multiple layers expect to learn the information of different scales, we design local multi-scale reconstruction, where the lower and upper layers reconstruct fine-scale and coarse-scale supervision signals respectively. This design not only accelerates the representation learning process by explicitly guiding multiple layers, but also facilitates multi-scale semantical understanding to the input. Extensive experiments show that with significantly less pre-training burden, our model achieves comparable or better performance on classification, detection and segmentation tasks than existing MIM models. Code is available with both MindSpore and PyTorch.

DETR-like models have significantly boosted the performance of detectors and even outperformed classical convolutional models. However, all tokens are treated equally without discrimination brings a redundant computational burden in the traditional encoder structure. The recent sparsification strategies exploit a subset of informative tokens to reduce attention complexity maintaining performance through the sparse encoder. But these methods tend to rely on unreliable model statistics. Moreover, simply reducing the token population hinders the detection performance to a large extent, limiting the application of these sparse models. We propose Focus-DETR, which focuses attention on more informative tokens for a better trade-off between computation efficiency and model accuracy. Specifically, we reconstruct the encoder with dual attention, which includes a token scoring mechanism that considers both localization and category semantic information of the objects from multi-scale feature maps. We efficiently abandon the background queries and enhance the semantic interaction of the fine-grained object queries based on the scores. Compared with the state-of-the-art sparse DETR-like detectors under the same setting, our Focus-DETR gets comparable complexity while achieving 50.4AP (+2.2) on COCO. The code is available at torch-version†and mindspore-version‡.

Recurrent neural network (RNNs) that are capable of modeling long-distance dependencies are widely used in various speech tasks, eg., keyword spotting (KWS) and speech enhancement (SE). Due to the limitation of power and memory in low-resource devices, efficient RNN models are urgently required for real-world applications. In this paper, we propose an efficient RNN architecture, GhostRNN, which reduces hidden state redundancy with cheap operations. In particular, we observe that partial dimensions of hidden states are similar to the others in trained RNN models, suggesting that redundancy exists in specific RNNs. To reduce the redundancy and hence computational cost, we propose to first generate a few intrinsic states, and then apply cheap operations to produce ghost states based on the intrinsic states. Experiments on KWS and SE tasks demonstrate that the proposed GhostRNN significantly reduces the memory usage (~40%) and computation cost while keeping performance similar.

At the heart of foundation models is the philosophy of "more is different", exemplified by the astonishing success in computer vision and natural language processing. However, the challenges of optimization and inherent complexity of transformer models call for a paradigm shift towards simplicity. In this study, we introduce VanillaNet, a neural network architecture that embraces elegance in design. By avoiding high depth, shortcuts, and intricate operations like self-attention, VanillaNet is refreshingly concise yet remarkably powerful. Each layer is carefully crafted to be compact and straightforward, with nonlinear activation functions pruned after training to restore the original architecture. VanillaNet overcomes the challenges of inherent complexity, making it ideal for resource-constrained environments. Its easy-to-understand and highly simplified architecture opens new possibilities for efficient deployment. Extensive experimentation demonstrates that VanillaNet delivers performance on par with renowned deep neural networks and vision transformers, showcasing the power of minimalism in deep learning. This visionary journey of VanillaNet has significant potential to redefine the landscape and challenge the status quo of foundation model, setting a new path for elegant and effective model design. Pre-trained models and codes are available at https://github.com/huawei-noah/VanillaNet and https://gitee.com/mindspore/models/tree/master/research/cv/vanillanet

The tremendous success of large models trained on extensive datasets demonstrates that scale is a key ingredient in achieving superior results. Therefore, the reflection on the rationality of designing knowledge distillation (KD) approaches for limited-capacity architectures solely based on small-scale datasets is now deemed imperative. In this paper, we identify the small data pitfall that presents in previous KD methods, which results in the underestimation of the power of vanilla KD framework on large-scale datasets such as ImageNet-1K. Specifically, we show that employing stronger data augmentation techniques and using larger datasets can directly decrease the gap between vanilla KD and other meticulously designed KD variants. This highlights the necessity of designing and evaluating KD approaches in the context of practical scenarios, casting off the limitations of small-scale datasets. Our investigation of the vanilla KD and its variants in more complex schemes, including stronger training strategies and different model capacities, demonstrates that vanilla KD is elegantly simple but astonishingly effective in large-scale scenarios. Without bells and whistles, we obtain state-of-the-art ResNet-50, ViT-S, and ConvNeXtV2-T models for ImageNet, which achieve 83.1%, 84.3%, and 85.0% top-1 accuracy, respectively. PyTorch code and checkpoints can be found at https://github.com/Hao840/vanillaKD.

Knowledge distillation (KD) has proven to be a highly effective approach for enhancing model performance through a teacher-student training scheme. However, most existing distillation methods are designed under the assumption that the teacher and student models belong to the same model family, particularly the hint-based approaches. By using centered kernel alignment (CKA) to compare the learned features between heterogeneous teacher and student models, we observe significant feature divergence. This divergence illustrates the ineffectiveness of previous hint-based methods in cross-architecture distillation. To tackle the challenge in distilling heterogeneous models, we propose a simple yet effective one-for-all KD framework called OFA-KD, which significantly improves the distillation performance between heterogeneous architectures. Specifically, we project intermediate features into an aligned latent space such as the logits space, where architecture-specific information is discarded. Additionally, we introduce an adaptive target enhancement scheme to prevent the student from being disturbed by irrelevant information. Extensive experiments with various architectures, including CNN, Transformer, and MLP, demonstrate the superiority of our OFA-KD framework in enabling distillation between heterogeneous architectures. Specifically, when equipped with our OFA-KD, the student models achieve notable performance improvements, with a maximum gain of 8.0% on the CIFAR-100 dataset and 0.7% on the ImageNet-1K dataset. PyTorch code and checkpoints can be found at https://github.com/Hao840/OFAKD.

The development of foundation vision models has pushed the general visual recognition to a high level, but cannot well address the fine-grained recognition in specialized domain such as invasive species classification. Identifying and managing invasive species has strong social and ecological value. Currently, most invasive species datasets are limited in scale and cover a narrow range of species, which restricts the development of deep-learning based invasion biometrics systems. To fill the gap of this area, we introduced Species196, a large-scale semi-supervised dataset of 196-category invasive species. It collects over 19K images with expert-level accurate annotations (Species196-L), and 1.2M unlabeled images of invasive species (Species196-U). The dataset provides four experimental settings for benchmarking the existing models and algorithms, namely, supervised learning, semi-supervised learning and self-supervised pretraining. To facilitate future research on these four learning paradigms, we conduct an empirical study of the representative methods on the introduced dataset. The dataset will be made publicly available at https://species-dataset.github.io/.

Source-free domain adaptive semantic segmentation aims to adapt a pre-trained source model to the unlabeled target domain without accessing the private source data. Previous methods usually fine-tune the entire network, which suffers from expensive parameter tuning. To avoid this problem, we propose to utilize visual prompt tuning for parameter-efficient adaptation. However, the existing visual prompt tuning methods are unsuitable for source-free domain adaptive semantic segmentation due to the following two reasons: (1) Commonly used visual prompts like input tokens or pixel-level perturbations cannot reliably learn informative knowledge beneficial for semantic segmentation. (2) Visual prompts require sufficient labeled data to fill the gap between the pre-trained model and downstream tasks. To alleviate these problems, we propose a universal unsupervised visual prompt tuning (Uni-UVPT) framework, which is applicable to various transformer-based backbones. Specifically, we first divide the source pre-trained backbone with frozen parameters into multiple stages, and propose a lightweight prompt adapter for progressively encoding informative knowledge into prompts and enhancing the generalization of target features between adjacent backbone stages. Cooperatively, a novel adaptive pseudo-label correction strategy with a multiscale consistency loss is designed to alleviate the negative effect of target samples with noisy pseudo labels and raise the capacity of visual prompts to spatial perturbations. Extensive experiments demonstrate that Uni-UVPT achieves state-of-the-art performance on GTA5 $\to$ Cityscapes and SYNTHIA $\to$ Cityscapes tasks and can serve as a universal and parameter-efficient framework for large-model unsupervised knowledge transfer. Code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/uni-uvpt and https://github.com/huawei-noah/noah-research/tree/master/uni-uvpt.

Convolutional Neural Networks (CNNs) are hard to deploy on edge devices due to its high computation and storage complexities. As a common practice for model compression, network pruning consists of two major categories: unstructured and structured pruning, where unstructured pruning constantly performs better. However, unstructured pruning presents a structured pattern at high pruning rates, which limits its performance. To this end, we propose a Rank-based PruninG (RPG) method to maintain the ranks of sparse weights in an adversarial manner. In each step, we minimize the low-rank approximation error for the weight matrices using singular value decomposition, and maximize their distance by pushing the weight matrices away from its low rank approximation. This rank-based optimization objective guides sparse weights towards a high-rank topology. The proposed method is conducted in a gradual pruning fashion to stabilize the change of rank during training. Experimental results on various datasets and different tasks demonstrate the effectiveness of our algorithm in high sparsity. The proposed RPG outperforms the state-of-the-art performance by 1.13\% top-1 accuracy on ImageNet in ResNet-50 with 98\% sparsity. The codes are available at https://github.com/huawei-noah/Efficient-Computing/tree/master/Pruning/RPG and https://gitee.com/mindspore/models/tree/master/research/cv/RPG.

Positive-Unlabeled (PU) learning aims to achieve high-accuracy binary classification with limited labeled positive examples and numerous unlabeled ones. Existing cost-sensitive-based methods often rely on strong assumptions that examples with an observed positive label were selected entirely at random. In fact, the uneven distribution of labels is prevalent in real-world PU problems, indicating that most actual positive and unlabeled data are subject to selection bias. In this paper, we propose a PU learning enhancement (PUe) algorithm based on causal inference theory, which employs normalized propensity scores and normalized inverse probability weighting (NIPW) techniques to reconstruct the loss function, thus obtaining a consistent, unbiased estimate of the classifier and enhancing the model's performance. Moreover, we investigate and propose a method for estimating propensity scores in deep learning using regularization techniques when the labeling mechanism is unknown. Our experiments on three benchmark datasets demonstrate the proposed PUe algorithm significantly improves the accuracy of classifiers on non-uniform label distribution datasets compared to advanced cost-sensitive PU methods. Codes are available at https://github.com/huawei-noah/Noah-research/tree/master/PUe and https://gitee.com/mindspore/models/tree/master/research/cv/PUe.

In the past years, YOLO-series models have emerged as the leading approaches in the area of real-time object detection. Many studies pushed up the baseline to a higher level by modifying the architecture, augmenting data and designing new losses. However, we find previous models still suffer from information fusion problem, although Feature Pyramid Network (FPN) and Path Aggregation Network (PANet) have alleviated this. Therefore, this study provides an advanced Gatherand-Distribute mechanism (GD) mechanism, which is realized with convolution and self-attention operations. This new designed model named as Gold-YOLO, which boosts the multi-scale feature fusion capabilities and achieves an ideal balance between latency and accuracy across all model scales. Additionally, we implement MAE-style pretraining in the YOLO-series for the first time, allowing YOLOseries models could be to benefit from unsupervised pretraining. Gold-YOLO-N attains an outstanding 39.9% AP on the COCO val2017 datasets and 1030 FPS on a T4 GPU, which outperforms the previous SOTA model YOLOv6-3.0-N with similar FPS by +2.4%. The PyTorch code is available at https://github.com/huawei-noah/Efficient-Computing/tree/master/Detection/Gold-YOLO, and the MindSpore code is available at https://gitee.com/mindspore/models/tree/master/research/cv/Gold_YOLO.

The extraordinary ability of generative models to generate photographic images has intensified concerns about the spread of disinformation, thereby leading to the demand for detectors capable of distinguishing between AI-generated fake images and real images. However, the lack of large datasets containing images from the most advanced image generators poses an obstacle to the development of such detectors. In this paper, we introduce the GenImage dataset, which has the following advantages: 1) Plenty of Images, including over one million pairs of AI-generated fake images and collected real images. 2) Rich Image Content, encompassing a broad range of image classes. 3) State-of-the-art Generators, synthesizing images with advanced diffusion models and GANs. The aforementioned advantages allow the detectors trained on GenImage to undergo a thorough evaluation and demonstrate strong applicability to diverse images. We conduct a comprehensive analysis of the dataset and propose two tasks for evaluating the detection method in resembling real-world scenarios. The cross-generator image classification task measures the performance of a detector trained on one generator when tested on the others. The degraded image classification task assesses the capability of the detectors in handling degraded images such as low-resolution, blurred, and compressed images. With the GenImage dataset, researchers can effectively expedite the development and evaluation of superior AI-generated image detectors in comparison to prevailing methodologies.

Transformer, first applied to the field of natural language processing, is a type of deep neural network mainly based on the self-attention mechanism. Thanks to its strong representation capabilities, researchers are looking at ways to apply transformer to computer vision tasks. In a variety of visual benchmarks, transformer-based models perform similar to or better than other types of networks such as convolutional and recurrent neural networks. Given its high performance and less need for vision-specific inductive bias, transformer is receiving more and more attention from the computer vision community. In this paper, we review these vision transformer models by categorizing them in different tasks and analyzing their advantages and disadvantages. The main categories we explore include the backbone network, high/mid-level vision, low-level vision, and video processing. We also include efficient transformer methods for pushing transformer into real device-based applications. Furthermore, we also take a brief look at the self-attention mechanism in computer vision, as it is the base component in transformer. Toward the end of this paper, we discuss the challenges and provide several further research directions for vision transformers.

This paper searches for the optimal neural architecture by minimizing a proxy of validation loss. Existing neural architecture search (NAS) methods used to discover the optimal neural architecture that best fits the validation examples given the up-to-date network weights. These intermediate validation results are invaluable but have not been fully explored. We propose to approximate the validation loss landscape by learning a mapping from neural architectures to their corresponding validate losses. The optimal neural architecture thus can be easily identified as the minimum of this proxy validation loss landscape. To improve the efficiency, a novel architecture sampling strategy is developed for the approximation of the proxy validation loss landscape. We also propose an operation importance weight (OIW) to balance the randomness and certainty of architecture sampling. The representation of neural architecture is learned through a graph autoencoder (GAE) over both architectures sampled during search and randomly generated architectures. We provide theoretical analyses on the validation loss estimator learned with our sampling strategy. Experimental results demonstrate that the proposed proxy validation loss landscape can be effective in both the differentiable NAS and the evolutionary-algorithm-based (EA-based) NAS.

To achieve efficient inference with a hardware-friendly design, Adder Neural Networks (ANNs) are proposed to replace expensive multiplication operations in Convolutional Neural Networks (CNNs) with cheap additions through utilizing ℓ1-norm for similarity measurement instead of cosine distance. However, we observe that there exists an increasing gap between CNNs and ANNs with reducing parameters, which cannot be eliminated by existing algorithms. In this paper, we present a simple yet effective Norm-Guided Distillation (NGD) method for ℓ1-norm ANNs to learn superior performance from ℓ2-norm ANNs. Although CNNs achieve similar accuracy with ℓ2-norm ANNs, the clustering performance based on ℓ2-distance can be easily learned by ℓ1-norm ANNs compared with cross correlation in CNNs. The features in ℓ2-norm ANNs are encouraged to achieve intra-class centralization and inter-class decentralization to amplify this advantage. Furthermore, the roughly estimated gradients in vanilla ANNs are modified to a progressive approximation from ℓ2-norm to ℓ1-norm so that a more accurate optimization can be achieved. Extensive evaluations on several benchmarks demonstrate the effectiveness of NGD on lightweight networks. For example, our method improves ANN by 10.43% with 0.25x GhostNet on CIFAR-100 and 3.1% with 1.0x GhostNet on ImageNet.

For the real-world time series analysis, data missing is a ubiquitously existing problem due to anomalies during data collecting and storage. If not treated properly, this problem will seriously hinder the classification, regression, or related tasks. Existing methods for time series imputation either impose too strong assumptions on the distribution of missing data or cannot fully exploit, even simply ignore, the informative temporal dependencies and feature correlations across different time steps. In this article, inspired by the idea of conditional generative adversarial networks, we propose a generative adversarial learning framework for time series imputation under the condition of observed data (as well as the labels, if possible). In our model, we employ a modified bidirectional RNN structure as the generator G, which is aimed at generating the missing values by taking advantage of the temporal and nontemporal information extracted from the observed time series. The discriminator D is designed to distinguish whether each value in a time series is generated or not so that it can help the generator to make an adjustment toward a more authentic imputation result. For an empirical verification of our model, we conduct imputation and classification experiments on several real-world time series data sets. The experimental results show an eminent improvement compared with state-of-the-art baseline models.

Many adaptations of transformers have emerged to address the single-modal vision tasks, where self-attention modules are stacked to handle input sources like images. Intuitively, feeding multiple modalities of data to vision transformers could improve the performance, yet the innermodal attentive weights may be diluted, which could thus greatly undermine the final performance. In this paper, we propose a multimodal token fusion method (TokenFusion), tailored for transformer-based vision tasks. To effectively fuse multiple modalities, TokenFusion dynamically detects uninformative tokens and substitute these tokens with projected and aggregated inter-modal features. Residual positional alignment is also adopted to enable explicit utilization of the inter-modal alignments after fusion. The design of TokenFusion allows the transformer to learn correlations among multimodal features, while the single-modal transformer architecture remains largely intact. Extensive experiments are conducted on a variety of homogeneous and heterogeneous modalities and demonstrate that TokenFusion surpasses state-of-the-art methods in three typical vision tasks: multimodal image-to-image translation, RGB-depth semantic segmentation, and 3D object detection with point cloud and images. Code will be released11https://github.com/huawei-noah/noah-research22https://gitee.com/mindspore/models/tree/master/research/cv/TokenFusion.

Quantization is an effective way to reduce the memory and computational costs of deep neural networks in which the full-precision weights and activations are represented using low-bit values. The bit-width for each layer in most of existing quantization methods is static, i.e., the same for all samples in the given dataset. However, natural images are of huge diversity with abundant content and using such a universal quantization configuration for all samples is not an optimal strategy. In this paper, we present to conduct the low-bit quantization for each image individually, and develop a dynamic quantization scheme for exploring their optimal bit-widths. To this end, a lightweight bit-controller is established and trained jointly with the given neural network to be quantized. During inference, the quantization configuration for an arbitrary image will be determined by the bit-widths generated by the controller, e.g., an image with simple texture will be allocated with lower bits and computational complexity and vice versa. Experimental results conducted on benchmarks demonstrate the effectiveness of the proposed dynamic quantization method for achieving state-of-art performance in terms of accuracy and computational complexity. The code will be available at https://github.com/huawei-noah/Efficient-Computing and https://gitee.com/mindspore/models/tree/master/research/cv/DynamicQuant.

Domain Adaptation aims to transfer the knowledge learned from a labeled source domain to an unlabeled target domain whose data distributions are different. However, the training data in source domain required by most of the existing methods is usually unavailable in real-world applications due to privacy preserving policies. Recently, Source-Free Domain Adaptation (SFDA) has drawn much attention, which tries to tackle domain adaptation problem without using source data. In this work, we propose a novel framework called SFDA-DE to address SFDA task via source Distribution Estimation. Firstly, we produce robust pseudo-labels for target data with spherical k-means clustering, whose initial class centers are the weight vectors (anchors) learned by the classifier of pretrained model. Furthermore, we propose to estimate the class-conditioned feature distribution of source domain by exploiting target data and corresponding anchors. Finally, we sample surrogate features from the estimated distribution, which are then utilized to align two domains by minimizing a contrastive adaptation loss function. Extensive experiments show that the proposed method achieves state-of-the-art performance on multiple DA benchmarks, and even outperforms traditional DA methods which require plenty of source data.

Vision transformers have been successfully applied to image recognition tasks due to their ability to capture long-range dependencies within an image. However, there are still gaps in both performance and computational cost between transformers and existing convolutional neural networks (CNNs). In this paper, we aim to address this issue and develop a network that can outperform not only the canonical transformers, but also the high-performance convolutional models. We propose a new transformer based hybrid network by taking advantage of transformers to capture long-range dependencies, and of CNNs to extract local information. Furthermore, we scale it to obtain a family of models, called CMTs, obtaining much better trade-off for accuracy and efficiency than previous CNN-based and transformer-based models. In particular, our CMT-S achieves 83.5% top-1 accuracy on ImageNet, while being 14x and 2x smaller on FLOPs than the existing DeiT and EfficientNet, respectively. The proposed CMT-S also generalizes well on CIFAR10 (99.2%), CIFAR100 (91.7%), Flowers (98.7%), and other challenging vision datasets such as COCO (44.3% mAP), with considerably less computational cost.

Previous vision MLPs such as MLP-Mixer and ResMLP accept linearly flattened image patches as input, making them inflexible for different input sizes and hard to capture spatial information. Such approach withholds MLPs from getting comparable performance with their transformer-based counterparts and prevents them from becoming a general backbone for computer vision. This paper presents Hire-MLP, a simple yet competitive vision MLP architecture via Hierarchical rearrangement, which contains two levels of rearrangements. Specifically, the inner-region rearrangement is proposed to capture local information inside a spatial region, and the cross-region rearrangement is proposed to enable information communication between different regions and capture global context by circularly shifting all tokens along spatial directions. Extensive experiments demonstrate the effectiveness of Hire-MLP as a versatile backbone for various vision tasks. In particular, Hire-MLP achieves competitive results on image classification, object detection and semantic segmentation tasks, e.g., 83.8% top-1 accuracy on ImageNet, 51.7% box AP and 44.8% mask AP on COCO val2017, and 49.9% mIoU on ADE20K, surpassing previous transformer-based and MLP-based models with better trade-off for accuracy and throughput.

Recently, Multilayer Perceptron (MLP) becomes the hotspot in the field of computer vision tasks. Without in-ductive bias, MLPs perform well on feature extraction and achieve amazing results. However, due to the simplic-ity of their structures, the performance highly depends on the local features communication machenism. To further improve the performance of MLP, we introduce information communication mechanisms from brain-inspired neu-ral networks. Spiking Neural Network (SNN) is the most famous brain-inspired neural network, and achieve great success on dealing with sparse data. Leaky Integrate and Fire (LIF) neurons in SNNs are used to communicate be-tween different time steps. In this paper, we incorporate the machanism of LIF neurons into the MLP models, to achieve better accuracy without extra FLOPs. We pro-pose a full-precision LIF operation to communicate be-tween patches, including horizontal LIF and vertical LIF in different directions. We also propose to use group LIF to extract better local features. With LIF modules, our SNN-MLP model achieves 81.9%, 83.3% and 83.5% top-1 accuracy on ImageNet dataset with only 4.4G, 8.5G and 15.2G FLOPs, respectively, which are state-of-the-art re-sults as far as we know. The source code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/snn_mlp.

This paper studies the efficiency problem for visual transformers by excavating redundant calculation in given networks. The recent transformer architecture has demonstrated its effectiveness for achieving excellent performance on a series of computer vision tasks. However, similar to that of convolutional neural networks, the huge computational cost of vision transformers is still a severe issue. Considering that the attention mechanism aggregates different patches layer-by-layer, we present a novel patch slimming approach that discards useless patches in a topdown paradigm. We first identify the effective patches in the last layer and then use them to guide the patch selection process of previous layers. For each layer, the impact of a patch on the final output feature is approximated and patches with less impacts will be removed. Experimental results on benchmark datasets demonstrate that the proposed method can significantly reduce the computational costs of vision transformers without affecting their performances. For example, over 45% FLOPs of the ViT-Ti model can be reduced with only 0.2% top-1 accuracy drop on the ImageNet dataset.

In the field of computer vision, recent works show that a pure MLP architecture mainly stacked by fully-connected layers can achieve competing performance with CNN and transformer. An input image of vision MLP is usually split into multiple tokens (patches), while the existing MLP models directly aggregate them with fixed weights, neglecting the varying semantic information of tokens from different images. To dynamically aggregate tokens, we propose to represent each token as a wave function with two parts, amplitude and phase. Amplitude is the original feature and the phase term is a complex value changing according to the semantic contents of input images. Introducing the phase term can dynamically modulate the relationship between tokens and fixed weights in MLP. Based on the wave-like token representation, we establish a novel Wave-MLP architecture for vision tasks. Extensive experiments demonstrate that the proposed Wave-MLP is superior to the state-of-the-art MLP architectures on various vision tasks such as image classification, object detection and semantic segmentation. The source code is available at https://github.com/huawei-noah/CV-Backbones/tree/master/wavemlp_pytorch and https://gitee.com/mindspore/models/tree/master/research/cv/wave_mlp.

This paper focuses on channel pruning for semantic segmen-tation networks. Previous methods to compress and acceler-ate deep neural networks in the classification task cannot be straightforwardly applied to the semantic segmentation net-work that involves an implicit multi-task learning problem via pre-training. To identify the redundancy in segmentation networks, we present a multi-task channel pruning approach. The importance of each convolution filter w.r.t. the channel of an arbitrary layer will be simultaneously determined by the classification and segmentation tasks. In addition, we de-velop an alternative scheme for optimizing importance scores of filters in the entire network. Experimental results on sev-eral benchmarks illustrate the superiority of the proposed al-gorithm over the state-of-the-art pruning methods. Notably, we can obtain an about 2 x FLOPs reduction on DeepLabv3 with only an about 1 % mIoU drop on the PASCAL VOC 2012 dataset and an about 1.3% mIoU drop on Cityscapes dataset, respectively.

This paper studies the 3D instance segmentation problem, which has a variety of real-world applications such as robotics and augmented reality. Since the surroundings of 3D objects are of high complexity, the separating of different objects is very difficult. To address this challenging problem, we propose a novel framework to group and refine the 3D instances. In practice, we first learn an offset vector for each point and shift it to its predicted instance center. To better group these points, we propose a Hierarchical Point Grouping algorithm to merge the centrally aggregated points progressively. All points are grouped into small clusters, which further gradually undergo another clustering procedure to merge into larger groups. These multi-scale groups are exploited for instance prediction, which is beneficial for predicting instances with different scales. In addition, a novel MaskScoreNet is developed to produce binary point masks of these groups for further refining the segmentation results. Extensive experiments conducted on the ScanNetV2 and S3DIS benchmarks demonstrate the effectiveness of the proposed method. For instance, our MaskGroup achieves a 66.4% mAP with the 0.5 IoU threshold on the ScanNetV2 test set, which is 1.9% higher than the state-of-the-art method.

Recently, neural architectures with all Multi-layer Perceptrons (MLPs) have attracted great research interest from the computer vision community. However, the inefficient mixing of spatial-channel information causes MLP-like vision models to demand tremendous pre-training on large-scale datasets. This work solves the problem from a novel knowledge distillation perspective. We propose a novel Spatial-channel Token Distillation (STD) method, which improves the information mixing in the two dimensions by introducing distillation tokens to each of them. A mutual information regularization is further introduced to let distillation tokens focus on their specific dimensions and maximize the performance gain. Extensive experiments on ImageNet for several MLP-like architectures demonstrate that the proposed token distillation mechanism can efficiently improve the accuracy. For example, the proposed STD boosts the top-1 accuracy of Mixer-S16 on ImageNet from 73.8% to 75.7% without any costly pre-training on JFT-300M. When applied to stronger architectures, e.g. CycleMLP-B1 and CycleMLP-B2, STD can still harvest about 1.1% and 0.5% accuracy gains, respectively.

We study the problem of learning from positive and unlabeled (PU) data in the federated setting, where each client only labels a little part of their dataset due to the limitation of resources and time. Different from the settings in traditional PU learning where the negative class consists of a single class, the negative samples which cannot be identified by a client in the federated setting may come from multiple classes which are unknown to the client. Therefore, existing PU learning methods can be hardly applied in this situation. To address this problem, we propose a novel framework, namely Federated learning with Positive and Unlabeled data (FedPU), to minimize the expected risk of multiple negative classes by leveraging the labeled data in other clients. We theoretically analyze the generalization bound of the proposed FedPU. Empirical experiments show that the FedPU can achieve much better performance than conventional supervised and semi-supervised federated learning methods.

Network architecture plays a key role in the deep learning-based computer vision system. The widely-used convolutional neural network and transformer treat the image as a grid or sequence structure, which is not flexible to capture irregular and complex objects. In this paper, we propose to represent the image as a graph structure and introduce a new \emph{Vision GNN} (ViG) architecture to extract graph-level feature for visual tasks. We first split the image to a number of patches which are viewed as nodes, and construct a graph by connecting the nearest neighbors. Based on the graph representation of images, we build our ViG model to transform and exchange information among all the nodes. ViG consists of two basic modules: Grapher module with graph convolution for aggregating and updating graph information, and FFN module with two linear layers for node feature transformation. Both isotropic and pyramid architectures of ViG are built with different model sizes. Extensive experiments on image recognition and object detection tasks demonstrate the superiority of our ViG architecture. We hope this pioneering study of GNN on general visual tasks will provide useful inspiration and experience for future research. The PyTorch code is available at \url{https://github.com/huawei-noah/Efficient-AI-Backbones} and the MindSpore code is available at \url{https://gitee.com/mindspore/models}.

Neural Architecture Search (NAS) can automatically design promising neural architectures without artificial experience. Though it achieves great success, prohibitively high search cost is required to find a high-performance architecture, which blocks its practical implementation. Neural predictor can directly evaluate the performance of neural networks based on their architectures and thereby save much budget. However, existing neural predictors require substantial annotated architectures trained from scratch, which still consume many computational resources. To solve this issue, we propose a Cross-Domain Predictor (CDP), which is trained based on the existing NAS benchmark datasets (e.g., NAS-Bench-101), but can be used to find high-performance architectures in large-scale search spaces. Particularly, we propose a progressive subspace adaptation strategy to address the domain discrepancy between the source architecture space and the target space. Considering the large difference between two architecture spaces, an assistant space is developed to smooth the transfer process. Compared with existing NAS methods, the proposed CDP is much more efficient. For example, CDP only requires the search cost of 0.1 GPU Days to find architectures with 76.9% top-1 accuracy on ImageNet and 97.51% on CIFAR-10.

The vulnerability of deep neural networks has been widely found in various models as well as tasks where slight perturbations on the inputs could lead to incorrect predictions. These perturbed inputs are known as adversarial examples and one of the intriguing properties of them is Adversarial Transfersability, i.e. the capability of adversarial examples to fool other models. Traditionally, this transferability is always regarded as a critical threat to the defense against adversarial attacks, however, we argue that the network robustness can be significantly boosted by utilizing adversarial transferability from a new perspective. In this work, we first discuss the influence of different popular normalization layers on the adversarial transferability, and then provide both empirical evidence and theoretical analysis to shed light on the relationship between normalization types and transferability. Based on our theoretical analysis, we propose a simple yet effective module named Random Normalization Aggregation (RNA) which replaces the batch normalization layers in the networks and aggregates different selected normalization types to form a huge random space. Specifically, a random path is sampled during each inference procedure so that the network itself can be treated as an ensemble of a wide range of different models. Since the entire random space is designed with low adversarial transferability, it is difficult to perform effective attacks even when the network parameters are accessible. We conduct extensive experiments on various models and datasets, and demonstrate the strong superiority of proposed algorithm. The PyTorch code is available at https://github.com/UniSerj/Random-Norm-Aggregation and the MindSpore code is available at https://gitee.com/mindspore/models/tree/master/research/cv/RNA.

In the past few years, transformers have achieved promising performance on various computer vision tasks. Unfortunately, the immense inference overhead of most existing vision transformers withholds them from being deployed on edge devices such as cell phones and smart watches. Knowledge distillation is a widely used paradigm for compressing cumbersome architectures into compact students via transferring information. However, most of them are designed for convolutional neural networks (CNNs), which do not fully investigate the character of vision transformers. In this paper, we fully utilize the patch-level information and propose a fine-grained manifold distillation method for transformer-based networks. Specifically, we train a tiny student model to match a pre-trained teacher model in the patch-level manifold space. Then, we decouple the manifold matching loss into three terms with careful design to further reduce the computational costs for the patch relationship. Equipped with the proposed method, a DeiT-Tiny model containing 5M parameters achieves 76.5\% top-1 accuracy on ImageNet-1k, which is +2.0\% higher than previous distillation approaches. Transfer learning results on other classification benchmarks and downstream vision tasks also demonstrate the superiority of our method over the state-of-the-art algorithms.

Transformer-based object detectors have shown competitive performance recently. Compared with convolutional neural networks limited by the relatively small receptive fields, the advantage of transformer for visual tasks is the capacity to perceive long-range dependencies among all image patches, while the deficiency is that the local fine-grained information is not fully excavated. In this paper, we introduce the Coarse-grained and Fine-grained crossing representations to build an efficient Detection Transformer (CFDT). Specifically, we propose a local-global cross fusion module to establish the connection between local fine-grained features and global coarse-grained features. Besides, we propose a coarse-fine aware neck which enables detection tokens to interact with both coarse-grained and fine-grained features. Furthermore, an efficient feature integration module is presented for fusing multi-scale representations from different stages. Experimental results on the COCO dataset demonstrate the effectiveness of the proposed method. For instance, our CFDT achieves 48.1 AP with 173G FLOPs, which possesses higher accuracy and less computation compared with the state-of-the-art transformer-based detector ViDT. Code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/CFDT.

Adder Neural Network (AdderNet) provides a new way for developing energy-efficient neural networks by replacing the expensive multiplications in convolution with cheaper additions (i.e., L1-norm). To achieve higher hardware efficiency, it is necessary to further study the low-bit quantization of AdderNet. Due to the limitation that the commutative law in multiplication does not hold in L1-norm, the well-established quantization methods on convolutional networks cannot be applied on AdderNets. Thus, the existing AdderNet quantization techniques propose to use only one shared scale to quantize both the weights and activations simultaneously. Admittedly, such an approach can keep the commutative law in the L1-norm quantization process, while the accuracy drop after low-bit quantization cannot be ignored. To this end, we first thoroughly analyze the difference on distributions of weights and activations in AdderNet and then propose a new quantization algorithm by redistributing the weights and the activations. Specifically, the pre-trained full-precision weights in different kernels are clustered into different groups, then the intra-group sharing and inter-group independent scales can be adopted. To further compensate the accuracy drop caused by the distribution difference, we then develop a lossless range clamp scheme for weights and a simple yet effective outliers clamp strategy for activations. Thus, the functionality of full-precision weights and the representation ability of full-precision activations can be fully preserved. The effectiveness of the proposed quantization method for AdderNet is well verified on several benchmarks, e.g., our 4-bit post-training quantized adder ResNet-18 achieves an 66.5% top-1 accuracy on the ImageNet with comparable energy efficiency, which is about 8.5% higher than that of the previous AdderNet quantization methods. Code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/AdderQuant.

Weight sparsity is a promising approach to reducing the model size and computation cost of convolutional neural networks (CNNs). Nevertheless, non-zero weights often distribute randomly in sparse CNN models, introducing enormous difficulty in obtaining actual speedup on common hardware (e.g., GPU) over their dense counterparts. Existing acceleration solutions either require hardware modifications for irregular memory access support or rely on a partially structured sparsity pattern. Neither of these methods is capable of achieving fruitful speedup on convolution layers.In this work, we propose an algorithm-software co-designed sparse convolution based on a novel out-vector-wise (OVW) sparse pattern. Building on the insight that vertical vector integrity can preserve continuous memory access in IM2COL, the OVW pattern treats a $V\times1$ vector as an entirety. To reduce the error caused by sparsity, we propose an equivalent transformation process, i.e., clustering-based channel permutation, to gather similar rows together. Experimental evaluations demonstrate that our method achieves a $1.7\times$ and $3.2\times$ speedup over the SOTA solution and the dense convolution of ResNet50 on NVIDIA V100 at 75\% sparsity, respectively, with only negligible accuracy loss. Moreover, compared to the SOTA solution that achieves speedups only on data with 60\% sparsity or more, our method begins to obtain speedups on data with only 10\% sparsity.

Light-weight convolutional neural networks (CNNs) are specially designed for applications on mobile devices with faster inference speed. The convolutional operation can only capture local information in a window region, which prevents performance from being further improved. Introducing self-attention into convolution can capture global information well, but it will largely encumber the actual speed. In this paper, we propose a hardware-friendly attention mechanism (dubbed DFC attention) and then present a new GhostNetV2 architecture for mobile applications. The proposed DFC attention is constructed based on fully-connected layers, which can not only execute fast on common hardware but also capture the dependence between long-range pixels. We further revisit the expressiveness bottleneck in previous GhostNet and propose to enhance expanded features produced by cheap operations with DFC attention, so that a GhostNetV2 block can aggregate local and long-range information simultaneously. Extensive experiments demonstrate the superiority of GhostNetV2 over existing architectures. For example, it achieves 75.3% top-1 accuracy on ImageNet with 167M FLOPs, significantly suppressing GhostNetV1 (74.5%) with a similar computational cost. The source code will be available at https://github.com/huawei-noah/Efficient-AI-Backbones/tree/master/ghostnetv2_pytorch and https://gitee.com/mindspore/models/tree/master/research/cv/ghostnetv2.

This paper studies the problem of designing compact binary architectures for vision multi-layer perceptrons (MLPs). We provide extensive analysis on the difficulty of binarizing vision MLPs and find that previous binarization methods perform poorly due to limited capacity of binary MLPs. In contrast with the traditional CNNs that utilizing convolutional operations with large kernel size, fully-connected (FC) layers in MLPs can be treated as convolutional layers with kernel size $1\times1$. Thus, the representation ability of the FC layers will be limited when being binarized, and places restrictions on the capability of spatial mixing and channel mixing on the intermediate features. To this end, we propose to improve the performance of binary MLP (BiMLP) model by enriching the representation ability of binary FC layers. We design a novel binary block that contains multiple branches to merge a series of outputs from the same stage, and also a universal shortcut connection that encourages the information flow from the previous stage. The downsampling layers are also carefully designed to reduce the computational complexity while maintaining the classification performance. Experimental results on benchmark dataset ImageNet-1k demonstrate the effectiveness of the proposed BiMLP models, which achieve state-of-the-art accuracy compared to prior binary CNNs.The MindSpore code is available at \url{https://gitee.com/mindspore/models/tree/master/research/cv/BiMLP}.

This paper introduces versatile filters to construct efficient convolutional neural networks that are widely used in various visual recognition tasks. Considering the demands of efficient deep learning techniques running on cost-effective hardware, a number of methods have been developed to learn compact neural networks. Most of these works aim to slim down filters in different ways, e.g., investigating small, sparse or quantized filters. In contrast, we treat filters from an additive perspective. A series of secondary filters can be derived from a primary filter with the help of binary masks. These secondary filters all inherit in the primary filter without occupying more storage, but once been unfolded in computation they could significantly enhance the capability of the filter by integrating information extracted from different receptive fields. Besides spatial versatile filters, we additionally investigate versatile filters from the channel perspective. Binary masks can be further customized for different primary filters under orthogonal constraints. We conduct theoretical analysis on network complexity and an efficient convolution scheme is introduced. Experimental results on benchmark datasets and neural networks demonstrate that our versatile filters are able to achieve comparable accuracy as that of original filters, but require less memory and computation cost.

Relying on the diverse graph convolution operations that have emerged in recent years, graph neural networks (GNNs) are shown to be powerful to deal with high-dimensional non-Euclidean domains, such as social networks or citation networks. Despite the tremendous human efforts been taken to explore new graph convolution operations, there are a few attempts to automatically search operations in GNNs. The search space of GNNs is significantly larger than that of CNNs, because of diverse components in the message-passing of GNNs. This, therefore, prevents the straightforward application of classical NAS methods for GNNs. In this work, we propose a novel dynamic one-shot search space for multi-branch neural architectures of GNNs. The dynamic search space maintains a subset of the large search space along with a set of importance weights for operation candidates in the subset as the architecture parameters. After each iteration, the subset is pruned by removing candidates with low importance weights and is expanded with new operations. The dynamic subsets of operation candidates are not uniform but is individual for each edge in the computation graph of the neural architecture, which can ensure the diversity of operations in the final architecture is as competitive as direct search in the large search space. Our experiments of semi-supervised and supervised node classification on citation networks, including Cora, Citeseer, and Pubmed, demonstrate that our method outperforms the current state-of-the-art manually designed architectures and reaches competitive performance to existing GNN NAS approaches with up to 10 times of speedup.

Despite the remarkable empirical performance of deep learning models, their vulnerability to adversarial examples has been revealed in many studies. They are prone to make a susceptible prediction to the input with imperceptible adversarial perturbation. Although recent works have remarkably improved the model's robustness under the adversarial training strategy, an evident gap between the natural accuracy and adversarial robustness inevitably exists. In order to mitigate this problem, in this paper, we assume that the robust and non-robust representations are two basic ingredients entangled in the integral representation. For achieving adversarial robustness, the robust representations of natural and adversarial examples should be disentangled from the non-robust part and the alignment of the robust representations can bridge the gap between accuracy and robustness. Inspired by this motivation, we propose a novel defense method called Deep Robust Representation Disentanglement Network (DRRDN). Specifically, DRRDN employs a disentangler to extract and align the robust representations from both adversarial and natural examples. Theoretical analysis guarantees the mitigation of the trade-off between robustness and accuracy with good disentanglement and alignment performance. Experimental results on benchmark datasets finally demonstrate the empirical superiority of our method.

As the computing power of modern hardware is increasing strongly, pre-trained deep learning models (e.g., BERT, GPT-3) learned on large-scale datasets have shown their effectiveness over conventional methods. The big progress is mainly contributed to the representation ability of transformer and its variant architectures. In this paper, we study the low-level computer vision task (e.g., denoising, super-resolution and deraining) and develop a new pre-trained model, namely, image processing transformer (IPT). To maximally excavate the capability of transformer, we present to utilize the well-known ImageNet benchmark for generating a large amount of corrupted image pairs. The IPT model is trained on these images with multi-heads and multi-tails. In addition, the contrastive learning is introduced for well adapting to different image processing tasks. The pre-trained model can therefore efficiently employed on desired task after fine-tuning. With only one pre-trained model, IPT outperforms the current state-of-the-art methods on various low-level benchmarks. Code is available at https://github.com/huawei-noah/Pretrained-IPT and https://gitee.com/mindspore/mindspore/tree/master/model_zoo/research/cv/IPT

Data-free learning for student networks is a new paradigm for solving users’ anxiety caused by the privacy problem of using original training data. Since the architectures of modern convolutional neural networks (CNNs) are compact and sophisticated, the alternative images or meta-data generated from the teacher network are often broken. Thus, the student network cannot achieve the comparable performance to that of the pre-trained teacher network especially on the large-scale image dataset. Different to previous works, we present to maximally utilize the massive available unlabeled data in the wild. Specifically, we first thoroughly analyze the output differences between teacher and student network on the original data and develop a data collection method. Then, a noisy knowledge distillation algorithm is proposed for achieving the performance of the student network. In practice, an adaptation matrix is learned with the student network for correcting the label noise produced by the teacher network on the collected unlabeled images. The effectiveness of our DFND (Data-Free Noisy Distillation) method is then verified on several benchmarks to demonstrate its superiority over state-of-the-art data-free distillation methods. Experiments on various datasets demonstrate that the student networks learned by the proposed method can achieve comparable performance with those using the original dataset. Code is available at https://github.com/huawei-noah/Data-Efficient-Model-Compression

Knowledge distillation is a widely used paradigm for inheriting information from a complicated teacher network to a compact student network and maintaining the strong performance. Different from image classification, object detectors are much more sophisticated with multiple loss functions in which features that semantic information rely on are tangled. In this paper, we point out that the information of features derived from regions excluding objects are also essential for distilling the student detector, which is usually ignored in existing approaches. In addition, we elucidate that features from different regions should be assigned with different importance during distillation. To this end, we present a novel distillation algorithm via decoupled features (DeFeat) for learning a better student detector. Specifically, two levels of decoupled features will be processed for embedding useful information into the student, i.e., decoupled features from neck and decoupled proposals from classification head. Extensive experiments on various detectors with different backbones show that the proposed DeFeat is able to surpass the state-of-the-art distillation methods for object detection. For example, DeFeat improves ResNet50 based Faster R-CNN from 37.4% to 40.9% mAP, and improves ResNet50 based RetinaNet from 36.5% to 39.7% mAP on COCO benchmark. Code will be released1,2.

Deep learning based object detection approaches have achieved great progress with the benefit from large amount of labeled images. However, image annotation remains a laborious, time-consuming and error-prone process. To further improve the performance of detectors, we seek to exploit all available labeled data and excavate useful samples from massive unlabeled images in the wild, which is rarely discussed before. In this paper, we present a positive-unlabeled learning based scheme to expand training data by purifying valuable images from massive unlabeled ones, where the original training data are viewed as positive data and the unlabeled images in the wild are unlabeled data. To effectively utilized these purified data, we propose a self-distillation algorithm based on hint learning and ground truth bounded knowledge distillation. Experimental results verify that the proposed positive-unlabeled data purification can strengthen the original detector by mining the massive unlabeled data. In particular, our method boosts the mAP of FPN by +2.0% on COCO benchmark.

This paper studies the single image super-resolution problem using adder neural networks (AdderNets). Com-pared with convolutional neural networks, AdderNets utilize additions to calculate the output features thus avoid massive energy consumptions of conventional multiplications. However, it is very hard to directly inherit the existing success of AdderNets on large-scale image classification to the image super-resolution task due to the different calculation paradigm. Specifically, the adder operation cannot easily learn the identity mapping, which is essential for image processing tasks. In addition, the functionality of high-pass filters cannot be ensured by AdderNets. To this end, we thoroughly analyze the relationship between an adder operation and the identity mapping and insert shortcuts to enhance the performance of SR models using adder networks. Then, we develop a learnable power activation for adjusting the feature distribution and refining details. Experiments conducted on several benchmark models and datasets demonstrate that, our image super-resolution models using AdderNets can achieve comparable performance and visual quality to that of their CNN baselines with an about 2.5× reduction on the energy consumption. The codes are available at: https://github.com/huawei-noah/AdderNet.

Neural network pruning is an essential approach for reducing the computational complexity of deep models so that they can be well deployed on resource-limited devices. Compared with conventional methods, the recently developed dynamic pruning methods determine redundant filters variant to each input instance which achieves higher acceleration. Most of the existing methods discover effective subnetworks for each instance independently and do not utilize the relationship between different inputs. To maximally excavate redundancy in the given network architecture, this paper proposes a new paradigm that dynamically removes redundant filters by embedding the manifold information of all instances into the space of pruned networks (dubbed as ManiDP). We first investigate the recognition complexity and feature similarity between images in the training set. Then, the manifold relationship between instances and the pruned sub-networks will be aligned in the training procedure. The effectiveness of the proposed method is verified on several benchmarks, which shows better performance in terms of both accuracy and computational cost compared to the state-of-the-art methods. For example, our method can reduce 55.3% FLOPs of ResNet-34 with only 0.57% top-1 accuracy degradation on ImageNet. The code will be available at https://github.com/huawei-noah/Pruning/tree/master/ManiDP.

An effective and efficient architecture performance evaluation scheme is essential for the success of Neural Architecture Search (NAS). To save computational cost, most of existing NAS algorithms often train and evaluate intermediate neural architectures on a small proxy dataset with limited training epochs. But it is difficult to expect an accurate performance estimation of an architecture in such a coarse evaluation way. This paper advocates a new neural architecture evaluation scheme, which aims to determine which architecture would perform better instead of accurately predict the absolute architecture performance. Therefore, we propose a relativistic architecture performance predictor in NAS (ReNAS). We encode neural architectures into feature tensors, and further refining the representations with the predictor. The proposed relativistic performance predictor can be deployed in discrete searching methods to search for the desired architectures without additional evaluation. Experimental results on NAS-Bench-101 dataset suggests that, sampling 424 (0.1% of the entire search space) neural architectures and their corresponding validation performance is already enough for learning an accurate architecture performance predictor. The accuracies of our searched neural architectures on NAS-Bench-101 and NAS-Bench-201 datasets are higher than that of the state-of-the-art methods and show the priority of the proposed method.

Neural Architecture Search (NAS) aims to automatically discover optimal architectures. In this paper, we propose an hourglass-inspired approach (HourNAS) for extremely fast NAS. It is motivated by the fact that the effects of the architecture often proceed from the vital few blocks. Acting like the narrow neck of an hourglass, vital blocks in the guaranteed path from the input to the output of a deep neural network restrict the information flow and influence the network accuracy. The other blocks occupy the major volume of the network and determine the overall network complexity, corresponding to the bulbs of an hourglass. To achieve an extremely fast NAS while preserving the high accuracy, we propose to identify the vital blocks and make them the priority in the architecture search. The search space of those non-vital blocks is further shrunk to only cover the candidates that are affordable under the computational resource constraints. Experimental results on ImageNet show that only using 3 hours (0.1 days) with one GPU, our HourNAS can search an architecture that achieves a 77.0% Top-1 accuracy, which outperforms the state-of-the-art methods.

Convolutional network compression methods require training data for achieving acceptable results, but training data is routinely unavailable due to some privacy and transmission limitations. Therefore, recent works focus on learning efficient networks without original training data, i.e., data-free model compression. Wherein, most of existing algorithms are developed for image recognition or segmentation tasks. In this paper, we study the data-free compression approach for single image super-resolution (SISR) task which is widely used in mobile phones and smart cameras. Specifically, we analyze the relationship between the outputs and inputs from the pre-trained network and explore a generator with a series of loss functions for maximally capturing useful information. The generator is then trained for synthesizing training samples which have similar distribution to that of the original data. To further alleviate the training difficulty of the student network using only the synthetic data, we introduce a progressive distillation scheme. Experiments on various datasets and architectures demonstrate that the pro-posed method is able to be utilized for effectively learning portable student networks without the original data, e.g., with 0.16dB PSNR drop on Set5 for ×2 super resolution. Code will be available at https://github.com/huawei-noah/Data-Efficient-Model-Compression.

Deep learning based methods, especially convolutional neural networks (CNNs) have been successfully applied in the field of single image super-resolution (SISR). To obtain better fidelity and visual quality, most of existing networks are of heavy design with massive computation. However, the computation resources of modern mobile devices are limited, which cannot easily support the expensive cost. To this end, this paper explores a novel frequency-aware dynamic network for dividing the input into multiple parts according to its coefficients in the discrete cosine transform (DCT) domain. In practice, the high-frequency part will be processed using expensive operations and the lower-frequency part is assigned with cheap operations to relieve the computation burden. Since pixels or image patches belong to low-frequency areas contain relatively few textural details, this dynamic network will not affect the quality of resulting super-resolution images. In addition, we embed predictors into the proposed dynamic network to end-to-end fine-tune the handcrafted frequency-aware masks. Extensive experiments conducted on benchmark SISR models and datasets show that the frequency-aware dynamic network can be employed for various SISR neural architectures to obtain the better tradeoff between visual quality and computational complexity. For instance, we can reduce the FLOPs of SR models by approximate 50% while preserving state-of-the-art SISR performance.

Adder neural network (AdderNet) is a new kind of deep model that replaces the original massive multiplications in convolutions by additions while preserving the high performance. Since the hardware complexity of additions is much lower than that of multiplications, the overall energy consumption is thus reduced significantly. To further optimize the hardware overhead of using AdderNet, this paper studies the winograd algorithm, which is a widely used fast algorithm for accelerating convolution and saving the computational costs. Unfortunately, the conventional Winograd algorithm cannot be directly applied to AdderNets since the distributive law in multiplication is not valid for the l1-norm. Therefore, we replace the element-wise multiplication in the Winograd equation by additions and then develop a new set of transform matrixes that can enhance the representation ability of output features to maintain the performance. Moreover, we propose the l2-to-l1 training strategy to mitigate the negative impacts caused by formal inconsistency. Experimental results on both FPGA and benchmarks show that the new method can further reduce the energy consumption without affecting the accuracy of the original AdderNet.

Adder neural networks (AdderNets) have shown impressive performance on image classification with only addition operations, which are more energy efficient than traditional convolutional neural networks built with multiplications. Compared with classification, there is a strong demand on reducing the energy consumption of modern object detectors via AdderNets for real-world applications such as autonomous driving and face detection. In this paper, we present an empirical study of AdderNets for object detection. We first reveal that the batch normalization statistics in the pre-trained adder backbone should not be frozen, since the relatively large feature variance of AdderNets. Moreover, we insert more shortcut connections in the neck part and design a new feature fusion architecture for avoiding the sparse features of adder layers. We present extensive ablation studies to explore several design choices of adder detectors. Comparisons with state-of-the-arts are conducted on COCO and PASCAL VOC benchmarks. Specifically, the proposed Adder FCOS achieves a 37.8% AP on the COCO val set, demonstrating comparable performance to that of the convolutional counterpart with an about $1.4\times$ energy reduction.

Adder neural network (AdderNet) replaces the original convolutions with massive multiplications by cheap additions while achieving comparable performance thus yields a series of energy-efficient neural networks. Compared with convolutional neural networks (CNNs), the training of AdderNets is much more sophisticated including several techniques for adjusting gradient and batch normalization. In addition, variances of both weights and activations in resulting adder networks are very enormous which limits its performance and the potential for applying to other tasks. To enhance the stability and robustness of AdderNets, we first thoroughly analyze the variance estimation of weight parameters and output features of an arbitrary adder layer. Then, we develop a weight normalization scheme for adaptively optimizing the weight distribution of AdderNets during the training procedure, which can reduce the perturbation on running mean and variance in batch normalization layers. Meanwhile, the proposed weight normalization can also be utilized to enhance the adversarial robustness of resulting networks. Experiments conducted on several benchmarks demonstrate the superiority of the proposed approach for generating AdderNets with higher performance.

Adder neural networks (ANNs) are designed for low energy cost which replace expensive multiplications in convolutional neural networks (CNNs) with cheaper additions to yield energy-efficient neural networks and hardware accelerations. Although ANNs achieve satisfactory efficiency, there exist gaps between ANNs and CNNs where the accuracy of ANNs can hardly be compared to CNNs without the assistance of other training tricks, such as knowledge distillation. The inherent discrepancy lies in the similarity measurement between filters and features, however how to alleviate this difference remains unexplored. To locate the potential problem of ANNs, we focus on the property difference due to similarity measurement. We demonstrate that unordered heavy tails in ANNs could be the key component which prevents ANNs from achieving superior classification performance since fatter tails tend to overlap in feature space. Through pre-defining Multivariate Skew Laplace distributions and embedding feature distributions into the loss function, ANN features can be fully controlled and designed for various properties. We further present a novel method for tackling existing heavy tails in ANNs with only a modification of classifier where ANN features are clustered with their tails well-formulated through proposed angle-based constraint on the distribution parameters to encourage high diversity of tails. Experiments conducted on several benchmarks and comparison with other distributions demonstrate the effectiveness of proposed approach for boosting the performance of ANNs.

Transformer is a new kind of neural architecture which encodes the input data as powerful features via the attention mechanism. Basically, the visual transformers first divide the input images into several local patches and then calculate both representations and their relationship. Since natural images are of high complexity with abundant detail and color information, the granularity of the patch dividing is not fine enough for excavating features of objects in different scales and locations. In this paper, we point out that the attention inside these local patches are also essential for building visual transformers with high performance and we explore a new architecture, namely, Transformer iN Transformer (TNT). Specifically, we regard the local patches (\eg, 16$\times$16) as “visual sentences” and present to further divide them into smaller patches (\eg, 4$\times$4) as “visual words”. The attention of each word will be calculated with other words in the given visual sentence with negligible computational costs. Features of both words and sentences will be aggregated to enhance the representation ability. Experiments on several benchmarks demonstrate the effectiveness of the proposed TNT architecture, \eg, we achieve an 81.5\% top-1 accuracy on the ImageNet, which is about 1.7\% higher than that of the state-of-the-art visual transformer with similar computational cost. The PyTorch code is available at \url{https://github.com/huawei-noah/CV-Backbones}, and the MindSpore code is available at \url{https://gitee.com/mindspore/models/tree/master/research/cv/TNT}.

With the tremendous advances in the architecture and scale of convolutional neural networks (CNNs) over the past few decades, they can easily reach or even exceed the performance of humans in certain tasks. However, a recently discovered shortcoming of CNNs is that they are vulnerable to adversarial attacks. Although the adversarial robustness of CNNs can be improved by adversarial training, there is a trade-off between standard accuracy and adversarial robustness. From the neural architecture perspective, this paper aims to improve the adversarial robustness of the backbone CNNs that have a satisfactory accuracy. Under a minimal computational overhead, the introduction of a dilation architecture is expected to be friendly with the standard performance of the backbone CNN while pursuing adversarial robustness. Theoretical analyses on the standard and adversarial error bounds naturally motivate the proposed neural architecture dilation algorithm. Experimental results on real-world datasets and benchmark neural networks demonstrate the effectiveness of the proposed algorithm to balance the accuracy and adversarial robustness.

Recently, transformer has achieved remarkable performance on a variety of computer vision applications. Compared with mainstream convolutional neural networks, vision transformers are often of sophisticated architectures for extracting powerful feature representations, which are more difficult to be developed on mobile devices. In this paper, we present an effective post-training quantization algorithm for reducing the memory storage and computational costs of vision transformers. Basically, the quantization task can be regarded as finding the optimal low-bit quantization intervals for weights and inputs, respectively. To preserve the functionality of the attention mechanism, we introduce a ranking loss into the conventional quantization objective that aims to keep the relative order of the self-attention results after quantization. Moreover, we thoroughly analyze the relationship between quantization loss of different layers and the feature diversity, and explore a mixed-precision quantization scheme by exploiting the nuclear norm of each attention map and output feature. The effectiveness of the proposed method is verified on several benchmark models and datasets, which outperforms the state-of-the-art post-training quantization algorithms. For instance, we can obtain an 81.29% top-1 accuracy using DeiT-B model on ImageNet dataset with about 8-bit quantization. Code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/VT-PTQ.

Transformer is a new kind of calculation paradigm for deep learning which has shown strong performance on a large variety of computer vision tasks. However, compared with conventional deep models (e.g., convolutional neural networks), vision transformers require more computational resources which cannot be easily deployed on mobile devices. To this end, we present to reduce the energy consumptions using adder neural network (AdderNet). We first theoretically analyze the mechanism of self-attention and the difficulty for applying adder operation into this module. Specifically, the feature diversity, i.e., the rank of attention map using only additions cannot be well preserved. Thus, we develop an adder attention layer that includes an additional identity mapping. With the new operation, vision transformers constructed using additions can also provide powerful feature representations. Experimental results on several benchmarks demonstrate that the proposed approach can achieve highly competitive performance to that of the baselines while achieving an about 2~3× reduction on the energy consumption.

Transformer models have achieved great progress on computer vision tasks recently. The rapid development of vision transformers is mainly contributed by their high representation ability for extracting informative features from input images. However, the mainstream transformer models are designed with deep architectures, and the feature diversity will be continuously reduced as the depth increases, \ie, feature collapse. In this paper, we theoretically analyze the feature collapse phenomenon and study the relationship between shortcuts and feature diversity in these transformer models. Then, we present an augmented shortcut scheme, which inserts additional paths with learnable parameters in parallel on the original shortcuts. To save the computational costs, we further explore an efficient approach that uses the block-circulant projection to implement augmented shortcuts. Extensive experiments conducted on benchmark datasets demonstrate the effectiveness of the proposed method, which brings about 1% accuracy increase of the state-of-the-art visual transformers without obviously increasing their parameters and FLOPs.

Binary neural networks (BNNs) represent original full-precision weights and activations into 1-bit with sign function. Since the gradient of the conventional sign function is almost zero everywhere which cannot be used for back-propagation, several attempts have been proposed to alleviate the optimization difficulty by using approximate gradient. However, those approximations corrupt the main direction of factual gradient. To this end, we propose to estimate the gradient of sign function in the Fourier frequency domain using the combination of sine functions for training BNNs, namely frequency domain approximation (FDA). The proposed approach does not affect the low-frequency information of the original sign function which occupies most of the overall energy, and high-frequency coefficients will be ignored to avoid the huge computational overhead. In addition, we embed a noise adaptation module into the training phase to compensate the approximation error. The experiments on several benchmark datasets and neural architectures illustrate that the binary network learned using our method achieves the state-of-the-art accuracy. Code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/FDA-BNN.

Deep convolutional neural networks (CNNs) are often of sophisticated design with numerous learnable parameters for the accuracy reason. To alleviate the expensive costs of deploying them on mobile devices, recent works have made huge efforts for excavating redundancy in pre-defined architectures. Nevertheless, the redundancy on the input resolution of modern CNNs has not been fully investigated, i.e., the resolution of input image is fixed. In this paper, we observe that the smallest resolution for accurately predicting the given image is different using the same neural network. To this end, we propose a novel dynamic-resolution network (DRNet) in which the input resolution is determined dynamically based on each input sample. Wherein, a resolution predictor with negligible computational costs is explored and optimized jointly with the desired network. Specifically, the predictor learns the smallest resolution that can retain and even exceed the original recognition accuracy for each image. During the inference, each input image will be resized to its predicted resolution for minimizing the overall computation burden. We then conduct extensive experiments on several benchmark networks and datasets. The results show that our DRNet can be embedded in any off-the-shelf network architecture to obtain a considerable reduction in computational complexity. For instance, DR-ResNet-50 achieves similar performance with an about 34% computation reduction, while gaining 1.4% accuracy increase with 10% computation reduction compared to the original ResNet-50 on ImageNet. Code will be available at https://gitee.com/mindspore/models/tree/master/research/cv/DRNet.

Deep convolutional neural networks have been widely used in numerous applications, but their demanding storage and computational resource requirements prevent their applications on mobile devices. Knowledge distillation aims to optimize a portable student network by taking the knowledge from a well-trained heavy teacher network. Traditional teacher-student-based methods used to rely on additional fully connected layers to bridge intermediate layers of teacher and student networks, which brings in a large number of auxiliary parameters. In contrast, this article aims to propagate information from teacher to student without introducing new variables that need to be optimized. We regard the teacher-student paradigm from a new perspective of feature embedding. By introducing the locality preserving loss, the student network is encouraged to generate the low-dimensional features that could inherit intrinsic properties of their corresponding high-dimensional features from the teacher network. The resulting portable network, thus, can naturally maintain the performance as that of the teacher network. Theoretical analysis is provided to justify the lower computation complexity of the proposed method. Experiments on benchmark data sets and well-trained networks suggest that the proposed algorithm is superior to state-of-the-art teacher-student learning methods in terms of computational and storage complexity.

Despite Generative Adversarial Networks (GANs) have been widely used in various image-to-image translation tasks, they can be hardly applied on mobile devices due to their heavy computation and storage cost. Traditional network compression methods focus on visually recognition tasks, but never deal with generation tasks. Inspired by knowledge distillation, a student generator of fewer parameters is trained by inheriting the low-level and high-level information from the original heavy teacher generator. To promote the capability of student generator, we include a student discriminator to measure the distances between real images, and images generated by student and teacher generators. An adversarial learning process is therefore established to optimize student generator and student discriminator. Qualitative and quantitative analysis by conducting experiments on benchmark datasets demonstrate that the proposed method can learn portable generative models with strong performance.

Although remarkable progress has been made on single image super-resolution due to the revival of deep convolutional neural networks, deep learning methods are confronted with the challenges of computation and memory consumption in practice, especially for mobile devices. Focusing on this issue, we propose an efficient residual dense block search algorithm with multiple objectives to hunt for fast, lightweight and accurate networks for image super-resolution. Firstly, to accelerate super-resolution network, we exploit the variation of feature scale adequately with the proposed efficient residual dense blocks. In the proposed evolutionary algorithm, the locations of pooling and upsampling operator are searched automatically. Secondly, network architecture is evolved with the guidance of block credits to acquire accurate super-resolution network. The block credit reflects the effect of current block and is earned during model evaluation process. It guides the evolution by weighing the sampling probability of mutation to favor admirable blocks. Extensive experimental results demonstrate the effectiveness of the proposed searching method and the found efficient super-resolution models achieve better performance than the state-of-the-art methods with limited number of parameters and FLOPs.

Deep neural networks often consist of a great number of trainable parameters for extracting powerful features from given datasets. One one hand, massive trainable parameters significantly enhance the performance of these deep networks. One the other hand, they bring the problem of over-fitting. To this end, dropout based methods disable some elements in the output feature maps during the training phase for reducing the co-adaptation of neurons. Although the generalization ability of the resulting models can be enhanced by these approaches, the conventional binary dropout is not the optimal solution. Therefore, we investigate the empirical Rademacher complexity related to intermediate layers of deep neural networks and propose a feature distortion method for addressing the aforementioned problem. In the training period, randomly selected elements in the feature maps will be replaced with specific values by exploiting the generalization error bound. The superiority of the proposed feature map distortion for producing deep neural network with higher testing performance is analyzed and demonstrated on several benchmark image datasets.

This paper studies the compression and acceleration of 3-dimensional convolutional neural networks (3D CNNs). To reduce the memory cost and computational complexity of deep neural networks, a number of algorithms have been explored by discovering redundant parameters in pre-trained networks. However, most of existing methods are designed for processing neural networks consisting of 2-dimensional convolution filters (i.e. image classification and detection) and cannot be straightforwardly applied for 3-dimensional filters (i.e. time series data). In this paper, we develop a novel approach for eliminating redundancy in the time dimensionality of 3D convolution filters by converting them into the frequency domain through a series of learned optimal transforms with extremely fewer parameters. Moreover, these transforms are forced to be orthogonal, and the calculation of feature maps can be accomplished in the frequency domain to achieve considerable speed-up rates. Experimental results on benchmark 3D CNN models and datasets demonstrate that the proposed Frequency Domain Compact 3D CNNs (FDC3D) can achieve the state-of-the-art performance, \eg a 2x speed-up ratio on the 3D-ResNet-18 without obviously affecting its accuracy.

Compared with cheap addition operation, multiplication operation is of much higher computation complexity. The widely-used convolutions in deep neural networks are exactly cross-correlation to measure the similarity between input feature and convolution filters, which involves massive multiplications between float values. In this paper, we present adder networks (AdderNets) to trade these massive multiplications in deep neural networks, especially convolutional neural networks (CNNs), for much cheaper additions to reduce computation costs. In AdderNets, we take the L1-norm distance between filters and input feature as the output response. The influence of this new similarity measure on the optimization of neural network have been thoroughly analyzed. To achieve a better performance, we develop a special back-propagation approach for AdderNets by investigating the full-precision gradient. We then propose an adaptive learning rate strategy to enhance the training procedure of AdderNets according to the magnitude of each neuron's gradient. As a result, the proposed AdderNets can achieve 74.9% Top-1 accuracy 91.7% Top-5 accuracy using ResNet-50 on the ImageNet dataset without any multiplication in convolutional layer. The codes are publicly available at: (https://github.com/huaweinoah/AdderNet).

Neural Architecture Search (NAS) has achieved great success in image classification task. Some recent works have managed to explore the automatic design of efficient backbone or feature fusion layer for object detection. However, these methods focus on searching only one certain component of object detector while leaving others manually designed. We identify the inconsistency between searched component and manually designed ones would withhold the detector of stronger performance. To this end, we propose a hierarchical trinity search framework to simultaneously discover efficient architectures for all components (i.e. backbone, neck, and head) of object detector in an end-to-end manner. In addition, we empirically reveal that different parts of the detector prefer different operators. Motivated by this, we employ a novel scheme to automatically screen different sub search spaces for different components so as to perform the end-to-end search for each component on the corresponding sub search space efficiently. Without bells and whistles, our searched architecture, namely Hit-Detector, achieves 41.4% mAP on COCO minival set with 27M parameters. Our implementation is available at https://github.com/ggjy/HitDet.pytorch.

This paper defines a positive and unlabeled classification problem for standard GANs, which then leads to a novel technique to stabilize the training of the discriminator in GANs. Traditionally, real data are taken as positive while generated data are negative. This positive-negative classification criterion was kept fixed all through the learning process of the discriminator without considering the gradually improved quality of generated data, even if they could be more realistic than real data at times. In contrast, it is more reasonable to treat the generated data as unlabeled, which could be positive or negative according to their quality. The discriminator is thus a classifier for this positive and unlabeled classification problem, and we derive a new Positive-Unlabeled GAN (PUGAN). We theoretically discuss the global optimality the proposed model will achieve and the equivalent optimization goal. Empirically, we find that PUGAN can achieve comparable or even better performance than those sophisticated discriminator stabilization methods.

Deploying convolutional neural networks (CNNs) on embedded devices is difficult due to the limited memory and computation resources. The redundancy in feature maps is an important characteristic of those successful CNNs, but has rarely been investigated in neural architecture design. This paper proposes a novel Ghost module to generate more feature maps from cheap operations. Based on a set of intrinsic feature maps, we apply a series of linear transformations with cheap cost to generate many ghost feature maps that could fully reveal information underlying intrinsic features. The proposed Ghost module can be taken as a plug-and-play component to upgrade existing convolutional neural networks. Ghost bottlenecks are designed to stack Ghost modules, and then the lightweight GhostNet can be easily established. Experiments conducted on benchmarks demonstrate that the proposed Ghost module is an impressive alternative of convolution layers in baseline models, and our GhostNet can achieve higher recognition performance (e.g. 75.7% top-1 accuracy) than MobileNetV3 with similar computational cost on the ImageNet ILSVRC-2012 classification dataset. Code is available at https://github.com/huawei-noah/ghostnet.

Neural architecture search (NAS) aims to automatically design deep neural networks of satisfactory performance. Wherein, architecture performance predictor is critical to efficiently value an intermediate neural architecture. But for the training of this predictor, a number of neural architectures and their corresponding real performance often have to be collected. In contrast with classical performance predictor optimized in a fully supervised way, this paper suggests a semi-supervised assessor of neural architectures. We employ an auto-encoder to discover meaningful representations of neural architectures. Taking each neural architecture as an individual instance in the search space, we construct a graph to capture their intrinsic similarities, where both labeled and unlabeled architectures are involved. A graph convolutional neural network is introduced to predict the performance of architectures based on the learned representations and their relation modeled by the graph. Extensive experimental results on the NAS-Benchmark-101 dataset demonstrated that our method is able to make a significant reduction on the required fully trained architectures for finding efficient architectures.

Searching techniques in most of existing neural architecture search (NAS) algorithms are mainly dominated by differentiable methods for the efficiency reason. In contrast, we develop an efficient continuous evolutionary approach for searching neural networks. Architectures in the population that share parameters within one SuperNet in the latest generation will be tuned over the training dataset with a few epochs. The searching in the next evolution generation will directly inherit both the SuperNet and the population, which accelerates the optimal network generation. The non-dominated sorting strategy is further applied to preserve only results on the Pareto front for accurately updating the SuperNet. Several neural networks with different model sizes and performances will be produced after the continuous search with only 0.4 GPU days. As a result, our framework provides a series of networks with the number of parameters ranging from 3.7M to 5.1M under mobile settings. These networks surpass those produced by the state-of-the-art methods on the benchmark ImageNet dataset.

This paper formalizes the binarization operations over neural networks from a learning perspective. In contrast to classical hand crafted rules (\eg hard thresholding) to binarize full-precision neurons, we propose to learn a mapping from full-precision neurons to the target binary ones. Each individual weight entry will not be binarized independently. Instead, they are taken as a whole to accomplish the binarization, just as they work together in generating convolution features. To help the training of the binarization mapping, the full-precision neurons after taking sign operations is regarded as some auxiliary supervision signal, which is noisy but still has valuable guidance. An unbiased estimator is therefore introduced to mitigate the influence of the supervision noise. Experimental results on benchmark datasets indicate that the proposed binarization technique attains consistent improvements over baselines.

This paper searches for the optimal neural architecture by minimizing a proxy of validation loss. Existing neural architecture search (NAS) methods used to discover the optimal neural architecture that best fits the validation examples given the up-to-date network weights. However, back propagation with a number of validation examples could be time consuming, especially when it needs to be repeated many times in NAS. Though these intermediate validation results are invaluable, they would be wasted if we cannot use them to predict the future from the past. In this paper, we propose to approximate the validation loss landscape by learning a mapping from neural architectures to their corresponding validate losses. The optimal neural architecture thus can be easily identified as the minimum of this proxy validation loss landscape. A novel sampling strategy is further developed for an efficient approximation of the loss landscape. Theoretical analysis indicates that the validation loss estimator learnt with our sampling strategy can reach a lower error rate and a lower label complexity compared with a uniform sampling. Experimental results on benchmarks demonstrate that the architecture searched by the proposed algorithm can achieve a satisfactory accuracy with less time cost.

Neural architecture search (NAS) has shown encouraging results in automating the architecture design. Recently, DARTS relaxes the search process with a differentiable formulation that leverages weight-sharing and SGD for cost reduction of NAS. In DARTS, all candidate operations are trained simultaneously during the network weight training step. Our empirical results show that this training procedure leads to the co-adaption problem and Matthew Effect: operations with fewer parameters would be trained maturely earlier. This causes two problems: firstly, the operations with more parameters may never have the chance to express the desired function since those with less have already done the job; secondly, the system will punish those underperforming operations by lowering their architecture parameter and backward smaller loss gradients, this causes the Matthew Effect. In this paper, we systematically study these problems and propose a novel grouped operation dropout algorithm named DropNAS to fix the problems with DARTS. Extensive experiments demonstrate that DropNAS solves the above issues and achieves promising performance. Specifically, DropNAS achieves 2.26% test error on CIFAR-10, 16.39% on CIFAR-100 and 23.4% on ImageNet (with the same training hyperparameters as DARTS for a fair comparison). It is also observed that DropNAS is robust across variants of the DARTS search space. Code is available at https://github.com/wiljohnhong/DropNAS.

Image compression, as one of the fundamental low-level image processing tasks, is very essential for computer vision. Tremendous computing and storage resources can be preserved with a trivial amount of visual information. Conventional image compression methods tend to obtain compressed images by minimizing their appearance discrepancy with the corresponding original images, but pay little attention to their efficacy in downstream perception tasks, e.g., image recognition and object detection. Thus, some of compressed images could be recognized with bias. In contrast, this paper aims to produce compressed images by pursuing both appearance and perceptual consistency. Based on the encoder-decoder framework, we propose using a pre-trained CNN to extract features of the original and compressed images, and making them similar. Thus the compressed images are discernible to subsequent tasks, and we name our method as Discernible Image Compression (DIC). In addition, the maximum mean discrepancy (MMD) is employed to minimize the difference between feature distributions. The resulting compression network can generate images with high image quality and preserve the consistent perception in the feature domain, so that these images can be well recognized by pre-trained machine learning models. Experiments on benchmarks demonstrate that images compressed by using the proposed method can also be well recognized by subsequent visual recognition and detection models. For instance, the mAP value of compressed images by DIC is about 0.6% higher than that of using compressed images by conventional methods.

To obtain excellent deep neural architectures, a series of techniques are carefully designed in EfficientNets. The giant formula for simultaneously enlarging the resolution, depth and width provides us a Rubik’s cube for neural networks. So that we can find networks with high efficiency and excellent performance by twisting the three dimensions. This paper aims to explore the twisting rules for obtaining deep neural networks with minimum model sizes and computational costs. Different from the network enlarging, we observe that resolution and depth are more important than width for tiny networks. Therefore, the original method, \ie the compound scaling in EfficientNet is no longer suitable. To this end, we summarize a tiny formula for downsizing neural architectures through a series of smaller models derived from the EfficientNet-B0 with the FLOPs constraint. Experimental results on the ImageNet benchmark illustrate that our TinyNet performs much better than the smaller version of EfficientNets using the inversed giant formula. For instance, our TinyNet-E achieves a 59.9\% Top-1 accuracy with only 24M FLOPs, which is about 1.9\% higher than that of the previous best MobileNetV3 with similar computational cost. Code will be available at \url{https://github.com/huawei-noah/CV-Backbones/tree/master/tinynet}, and \url{https://gitee.com/mindspore/mindspore/tree/master/model_zoo/research/cv/tinynet}.

Neural architecture search (NAS) has demonstrated impressive performance in automatically designing high-performance neural networks. The power of deep neural networks is to be unleashed for analyzing a large volume of data (e.g. ImageNet), but the architecture search is often executed on another smaller dataset (e.g. CIFAR-10) to finish it in a feasible time. However, it is hard to guarantee that the optimal architecture derived on the proxy task could maintain its advantages on another more challenging dataset. This paper aims to improve the generalization of neural architectures via domain adaptation. We analyze the generalization bounds of the derived architecture and suggest its close relations with the validation error and the data distribution distance on both domains. These theoretical analyses lead to AdaptNAS, a novel and principled approach to adapt neural architectures between domains in NAS. Our experimental evaluation shows that only a small part of ImageNet will be sufficient for AdaptNAS to extend its architecture success to the entire ImageNet and outperform state-of-the-art comparison algorithms.

By transferring both features and gradients between different layers, shortcut connections explored by ResNets allow us to effectively train very deep neural networks up to hundreds of layers. However, the additional computation costs induced by those shortcuts are often overlooked. For example, during online inference, the shortcuts in ResNet-50 account for about 40 percent of the entire memory usage on feature maps, because the features in the preceding layers cannot be released until the subsequent calculation is completed. In this work, for the first time, we consider training the CNN models with shortcuts and deploying them without. In particular, we propose a novel joint-training framework to train plain CNN by leveraging the gradients of the ResNet counterpart. During forward step, the feature maps of the early stages of plain CNN are passed through later stages of both itself and the ResNet counterpart to calculate the loss. During backpropagation, gradients calculated from a mixture of these two parts are used to update the plainCNN network to solve the gradient vanishing problem. Extensive experiments on ImageNet/CIFAR10/CIFAR100 demonstrate that the plainCNN network without shortcuts generated by our approach can achieve the same level of accuracy as that of the ResNet baseline while achieving about $1.4\times $ speed-up and $1.25\times$ memory reduction. We also verified the feature transferability of our ImageNet pretrained plain-CNN network by fine-tuning it on MIT 67 and Caltech 101. Our results show that the performance of the plain-CNN is slightly higher than that of its baseline ResNet-50 on these two datasets. The codes are in: \href{https://github.com/leoozy/JointRD_Neurips2020}{https://github.com/leoozy/JointRD\_Neurips2020}

This paper proposes a reliable neural network pruning algorithm by setting up a scientific control. Existing pruning methods have developed various hypotheses to approximate the importance of filters to the network and then execute filter pruning accordingly. To increase the reliability of the results, we prefer to have a more rigorous research design by including a scientific control group as an essential part to minimize the effect of all factors except the association between the filter and expected network output. Acting as a control group, knockoff feature is generated to mimic the feature map produced by the network filter, but they are conditionally independent of the example label given the real feature map. We theoretically suggest that the knockoff condition can be approximately preserved given the information propagation of network layers. Besides the real feature map on an intermediate layer, the corresponding knockoff feature is brought in as another auxiliary input signal for the subsequent layers. Redundant filters can be discovered in the adversarial process of different features. Through experiments, we demonstrate the superiority of the proposed algorithm over state-of-the-art methods. For example, our method can reduce 57.8% parameters and 60.2% FLOPs of ResNet-101 with only 0.01% top-1 accuracy loss on ImageNet.

Adder Neural Networks (ANNs) which only contain additions bring us a new way of developing deep neural networks with low energy consumption. Unfortunately, there is an accuracy drop when replacing all convolution filters by adder filters. The main reason here is the optimization difficulty of ANNs using $\ell_1$-norm, in which the estimation of gradient in back propagation is inaccurate. In this paper, we present a novel method for further improving the performance of ANNs without increasing the trainable parameters via a progressive kernel based knowledge distillation (PKKD) method. A convolutional neural network (CNN) with the same architecture is simultaneously initialized and trained as a teacher network, features and weights of ANN and CNN will be transformed to a new space to eliminate the accuracy drop. The similarity is conducted in a higher-dimensional space to disentangle the difference of their distributions using a kernel based method. Finally, the desired ANN is learned based on the information from both the ground-truth and teacher, progressively. The effectiveness of the proposed method for learning ANN with higher performance is then well-verified on several benchmarks. For instance, the ANN-50 trained using the proposed PKKD method obtains a 76.8\% top-1 accuracy on ImageNet dataset, which is 0.6\% higher than that of the ResNet-50.

Quantized neural networks with low-bit weights and activations are attractive for developing AI accelerators. However, the quantization functions used in most conventional quantization methods are non-differentiable, which increases the optimization difficulty of quantized networks. Compared with full-precision parameters (\emph{i.e.}, 32-bit floating numbers), low-bit values are selected from a much smaller set. For example, there are only 16 possibilities in 4-bit space. Thus, we present to regard the discrete weights in an arbitrary quantized neural network as searchable variables, and utilize a differential method to search them accurately. In particular, each weight is represented as a probability distribution over the discrete value set. The probabilities are optimized during training and the values with the highest probability are selected to establish the desired quantized network. Experimental results on benchmarks demonstrate that the proposed method is able to produce quantized neural networks with higher performance over the state-of-the-arts on both image classification and super-resolution tasks.

Here we study the low-resolution visual recognition problem. Conventional methods are usually trained on images with large ROIs (regions of interest), while the regions and insider images are often small and blur in real-world applications. Therefore, deep neural networks learned on high-resolution images cannot be directly used for recognizing low-resolution objects. To overcome this challenging problem, we propose to use the teacher-student learning paradigm for distilling useful feature information from a pre-trained deep model on high-resolution visual data. In practice, a distillation loss is used to seek the perceptual consistency of low-resolution images and high-resolution images. By simultaneously optimizing the recognition loss and distillation loss, we formulate a novel low-resolution recognition approach. Experiments conducted on benchmarks demonstrate that the proposed method is capable to learn well-performed models for recognizing low-resolution objects, which is superior to the state-of-the-art methods.

Learning portable neural networks is very essential for computer vision for the purpose that pre-trained heavy deep models can be well applied on edge devices such as mobile phones and micro sensors. Most existing deep neural network compression and speed-up methods are very effective for training compact deep models, when we can directly access the training dataset. However, training data for the given deep network are often unavailable due to some practice problems (\eg privacy, legal issue, and transmission), and the architecture of the given network are also unknown except some interfaces. To this end, we propose a novel framework for training efficient deep neural networks by exploiting generative adversarial networks (GANs). To be specific, the pre-trained teacher networks are regarded as a fixed discriminator and the generator is utilized for derivating training samples which can obtain the maximum response on the discriminator. Then, an efficient network with smaller model size and computational complexity is trained using the generated data and the teacher network, simultaneously. Efficient student networks learned using the proposed Data-Free Learning (DFL) method achieve 92.22% and 74.47% accuracies without any training data on the CIFAR-10 and CIFAR-100 datasets, respectively. Meanwhile, our student network obtains an 80.56% accuracy on the CelebA benchmark.

Generative adversarial networks (GANs) have been successfully used for considerable computer vision tasks, especially the image-to-image translation. However, generators in these networks are of complicated architectures with large number of parameters and huge computational complexities. Existing methods are mainly designed for compressing and speeding-up deep neural networks in the classification task, and cannot be directly applied on GANs for image translation, due to their different objectives and training procedures. To this end, we develop a novel co-evolutionary approach for reducing their memory usage and FLOPs simultaneously. In practice, generators for two image domains are encoded as two populations and synergistically optimized for investigating the most important convolution filters iteratively. Fitness of each individual is calculated using the number of parameters, a discriminator-aware regularization, and the cycle consistency. Extensive experiments conducted on benchmark datasets demonstrate the effectiveness of the proposed method for obtaining compact and effective generators.

This paper aims to build efficient convolutional neural networks using a set of Lego filters. Many successful building blocks, e.g., inception and residual modules, have been designed to refresh state-of-the-art records of CNNs on visual recognition tasks. Beyond these high-level modules, we suggest that an ordinary filter in the neural network can be upgraded to a sophisticated module as well. Filter modules are established by assembling a shared set of Lego filters that are often of much lower dimensions. Weights in Lego filters and binary masks to stack Lego filters for these filter modules can be simultaneously optimized in an end-to-end manner as usual. Inspired by network engineering, we develop a split-transform-merge strategy for an efficient convolution by exploiting intermediate Lego feature maps. The compression and acceleration achieved by Lego Networks using the proposed Lego filters have been theoretically discussed. Experimental results on benchmark datasets and deep models demonstrate the advantages of the proposed Lego filters and their potential real-world applications on mobile devices.

Network pruning is widely applied to deep CNN models due to their heavy computation costs and achieves high performance by keeping important weights while removing the redundancy. Pruning redundant weights directly may hurt global information flow, which suggests that an efficient sparse network should take graph properties into account. Thus, instead of paying more attention to preserving important weight, we focus on the pruned architecture itself. We propose to use graph entropy as the measurement, which shows useful properties to craft high-quality neural graphs and enables us to propose efficient algorithm to construct them as the initial network architecture. Our algorithm can be easily implemented and deployed to different popular CNN models and achieve better trade-offs.

This paper expands the strength of deep convolutional neural networks (CNNs) to the pedestrian attribute recognition problem by devising a novel attribute aware pooling algorithm. Existing vanilla CNNs cannot be straightforwardly applied to handle multi-attribute data because of the larger label space as well as the attribute entanglement and correlations. We tackle these challenges that hampers the development of CNNs for multi-attribute classification by fully exploiting the correlation between different attributes. The multi-branch architecture is adopted for fucusing on attributes at different regions. Besides the prediction based on each branch itself, context information of each branch are employed for decision as well. The attribute aware pooling is developed to integrate both kinds of information. Therefore, attributes which are indistinct or tangled with others can be accurately recognized by exploiting the context information. Experiments on benchmark datasets demonstrate that the proposed pooling method appropriately explores and exploits the correlations between attributes for the pedestrian attribute recognition.

Exploring deep convolutional neural networks of high efficiency and low memory usage is very essential for a wide variety of machine learning tasks. Most of existing approaches used to accelerate deep models by manipulating parameters or filters without data, e.g., pruning and decomposition. In contrast, we study this problem from a different perspective by respecting the difference between data. An instance-wise feature pruning is developed by identifying informative features for different instances. Specifically, by investigating a feature decay regularization, we expect intermediate feature maps of each instance in deep neural networks to be sparse while preserving the overall network performance. During online inference, subtle features of input images extracted by intermediate layers of a well-trained neural network can be eliminated to accelerate the subsequent calculations. We further take coefficient of variation as a measure to select the layers that are appropriate for acceleration. Extensive experiments conducted on benchmark datasets and networks demonstrate the effectiveness of the proposed method.

Many attempts have been done to extend the great success of convolutional neural networks (CNNs) achieved on high-end GPU servers to portable devices such as smart phones. Providing compression and acceleration service of deep learning models on the cloud is therefore of significance and is attractive for end users. However, existing network compression and acceleration approaches usually fine-tuning the svelte model by requesting the entire original training data (e.g. ImageNet), which could be more cumbersome than the network itself and cannot be easily uploaded to the cloud. In this paper, we present a novel positive-unlabeled (PU) setting for addressing this problem. In practice, only a small portion of the original training set is required as positive examples and more useful training examples can be obtained from the massive unlabeled data on the cloud through a PU classifier with an attention based multi-scale feature extractor. We further introduce a robust knowledge distillation (RKD) scheme to deal with the class imbalance problem of these newly augmented training examples. The superiority of the proposed method is verified through experiments conducted on the benchmark models and datasets. We can use only 8% of uniformly selected data from the ImageNet to obtain an efficient model with comparable performance to the baseline ResNet-34.

Deep convolutional neural networks (CNNs) are successfully used in a number of applications. However, their storage and computational requirements have largely prevented their widespread use on mobile devices. Here we present a series of approaches for compressing and speeding up CNNs in the frequency domain, which focuses not only on smaller weights but on all the weights and their underlying connections. By treating convolution filters as images, we decompose their representations in the frequency domain as common parts (i.e., cluster centers) shared by other similar filters and their individual private parts (i.e., individual residuals). A large number of low-energy frequency coefficients in both parts can be discarded to produce high compression without significantly compression romising accuracy. Furthermore, we explore a data-driven method for removing redundancies in both spatial and frequency domains, which allows us to discard more useless weights by keeping similar accuracies. After obtaining the optimal sparse CNN in the frequency domain, we relax the computational burden of convolution operations in CNNs by linearly combining the convolution responses of discrete cosine transform (DCT) bases. The compression and speed-up ratios of the proposed algorithm are thoroughly analyzed and evaluated on benchmark image datasets to demonstrate its superiority over state-of-the-art methods.

Effective methods for learning deep neural networks with fewer parameters are urgently required, since storage and computations of heavy neural networks have largely prevented their widespread use on mobile devices. Compared with algorithms which directly remove weights or filters for obtaining considerable compression and speed-up ratios, training thin deep networks exploiting the student-teacher learning paradigm is more flexible. However, it is very hard to determine which formulation is optimal to measure the information inherited from teacher networks. To overcome this challenge, we utilize the generative adversarial network (GAN) to learn the student network. In practice, the generator is exactly the student network with extremely less parameters and the discriminator is used as a teaching assistant for distinguishing features extracted from student and teacher networks. By simultaneously optimizing the generator and the discriminator, the resulting student network can produce features of input data with the similar distribution as that of features of the teacher network. Extensive experimental results on benchmark datasets demonstrate that the proposed method is capable of learning well-performed portable networks, which is superior to the state-of-the-art methods.

High-dimensional data in many areas such as computer vision and machine learning tasks brings in computational and analytical difficulty. Feature selection which selects a subset from observed features is a widely used approach for improving performance and effectiveness of machine learning models with high-dimensional data. In this paper, we propose a novel AutoEncoder Feature Selector (AEFS) for unsupervised feature selection which combines autoencoder regression and group lasso tasks. Compared to traditional feature selection methods, AEFS can select the most important features by excavating both linear and nonlinear information among features, which is more flexible than the conventional self-representation method for unsupervised feature selection with only linear assumptions. Experimental results on benchmark dataset show that the proposed method is superior to the state-of-the-art method.

Compressing convolutional neural networks (CNNs) is essential for transferring the success of CNNs to a wide variety of applications to mobile devices. In contrast to directly recognizing subtle weights or filters as redundant in a given CNN, this paper presents an evolutionary method to automatically eliminate redundant convolution filters. We represent each compressed network as a binary individual of specific fitness. Then, the population is upgraded at each evolutionary iteration using genetic operations. As a result, an extremely compact CNN is generated using the fittest individual, which has the original network structure and can be directly deployed in any off-the-shelf deep learning libraries. In this approach, either large or small convolution filters can be redundant, and filters in the compressed network are more distinct. In addition, since the number of filters in each convolutional layer is reduced, the number of filter channels and the size of feature maps are also decreased, naturally improving both the compression and speed-up ratios. Experiments on benchmark deep CNN models suggest the superiority of the proposed algorithm over the state-of-the-art compression methods, e.g. combined with the parameter refining approach, we can reduce the storage requirement and the floating-point multiplications of ResNet-50 by a factor of 14.64x and 5.19x, respectively, without affecting its accuracy.

This paper introduces versatile filters to construct efficient convolutional neural network. Considering the demands of efficient deep learning techniques running on cost-effective hardware, a number of methods have been developed to learn compact neural networks. Most of these works aim to slim down filters in different ways, e.g., investigating small, sparse or binarized filters. In contrast, we treat filters from an additive perspective. A series of secondary filters can be derived from a primary filter. These secondary filters all inherit in the primary filter without occupying more storage, but once been unfolded in computation they could significantly enhance the capability of the filter by integrating information extracted from different receptive fields. Besides spatial versatile filters, we additionally investigate versatile filters from the channel perspective. The new techniques are general to upgrade filters in existing CNNs. Experimental results on benchmark datasets and neural networks demonstrate that CNNs constructed with our versatile filters are able to achieve comparable accuracy as that of original filters, but require less memory and FLOPs.

Discovering robust low-rank data representations is important in many real-world problems. Traditional robust principal component analysis (RPCA) assumes that the observed data are corrupted by some sparse noise (e.g., Laplacian noise) and utilizes the l1-norm to separate out the noisy compo- nent. Nevertheless, as well as simple Gaussian or Laplacian noise, noise in real-world data is often more complex, and thus the l1 and l2-norms are insufficient for noise charac- terization. This paper presents a more flexible approach to modeling complex noise by investigating their properties in the frequency domain. Although elements of a noise matrix are chaotic in the spatial domain, the absolute values of its alternative coefficients in the frequency domain are constant w.r.t. their variance. Based on this observation, a new robust PCA algorithm is formulated by simultaneously discovering the low-rank and noisy components. Extensive experiments on synthetic data and video background subtraction demon- strate that FRPCA is effective for handles complex noise.

Convolutional neural networks (CNNs) have shown extraordinary performance in a number of applications, but they are usually of heavy design for the accuracy reason. Beyond compressing the filters in CNNs, this paper focuses on the redundancy in the feature maps derived from the large number of filters in a layer. We propose to extract intrinsic representation of the feature maps and preserve the discriminability of the features. Circulant matrix is employed to formulate the feature map transformation, which only requires O(dlog d) computation complexity to embed a d-dimensional feature map. The filter is then re-configured to establish the mapping from original input to the new compact feature map, and the resulting network can preserve intrinsic information of the original network with significantly fewer parameters, which not only decreases the online memory for launching CNN but also accelerates the computation speed. Experiments on benchmark image datasets demonstrate the superiority of the proposed algorithm over state-of-the-art methods.

This paper presents privileged multi-label learning (PrML) to explore and exploit the relationship between labels in multi-label learning problems. We suggest that for each individual label, it cannot only be implicitly connected with other labels via the low-rank constraint over label predictors, but also its performance on examples can receive the explicit comments from other labels together acting as an Oracle teacher. We generate privileged label feature for each example and its individual label, and then integrate it into the framework of low-rank based multi-label learning. The proposed algorithm can therefore comprehensively explore and exploit label relationships by inheriting all the merits of privileged information and low-rank constraints. We show that PrML can be efficiently solved by dual coordinate descent algorithm using iterative optimization strategy with cheap updates. Experiments on benchmark datasets show that through privileged label features, the performance can be significantly improved and PrML is superior to several competing methods in most cases.

Here we study the extreme visual recovery problem, in which over 90% of pixel values in a given image are missing. Existing low rank-based algorithms are only effective for recovering data with at most 90% missing values. Thus, we exploit visual data's smoothness property to help solve this challenging extreme visual recovery problem. Based on the discrete cosine transform (DCT), we propose a novel DCT regularizer that involves all pixels and produces smooth estimations in any view. Our theoretical analysis shows that the total variation regularizer, which only achieves local smoothness, is a special case of the proposed DCT regularizer. We also develop a new visual recovery algorithm by minimizing the DCT regularizer and nuclear norm to achieve a more visually pleasing estimation. Experimental results on a benchmark image data set demonstrate that the proposed approach is superior to the state-of-the-art methods in terms of peak signal-to-noise ratio and structural similarity.

Deep convolutional neural networks (CNNs) are successfully used in a number of applications. However, their storage and computational requirements have largely prevented their widespread use on mobile devices. Here we present an effective CNN compression approach in the frequency domain, which focuses not only on smaller weights but on all the weights and their underlying connections. By treating convolutional filters as images, we decompose their representations in the frequency domain as common parts (i.e., cluster centers) shared by other similar filters and their individual private parts (i.e., individual residuals). A large number of low-energy frequency coefficients in both parts can be discarded to produce high compression without significantly compromising accuracy. We relax the computational burden of convolution operations in CNNs by linearly combining the convolution responses of discrete cosine transform (DCT) bases. The compression and speed-up ratios of the proposed algorithm are thoroughly analyzed and evaluated on benchmark image datasets to demonstrate its superiority over state-of-the-art methods.

Scale invariant feature transform (SIFT) is effective for representing images in computer vision tasks, as one of the most resistant feature descriptions to common image deformations. However, two issues should be addressed: first, feature description based on gradient accumulation is not compact and contains redundancies; second, multiple orientations are often extracted from one local region and therefore produce multiple descriptions, which is not good for memory efficiency. To resolve these two issues, this paper introduces a novel method to determine the dominant orientation for multiple-orientation cases, named discrete cosine transform (DCT) intrinsic orientation, and a new DCT inspired feature transform (DIFT). In each local region, it first computes a unique DCT intrinsic orientation via DCT matrix and rotates the region accordingly, and then describes the rotated region with partial DCT matrix coefficients to produce an optimized low-dimensional descriptor. We test the accuracy and robustness of DIFT on real image matching. Afterward, extensive applications performed on public benchmarks for visual retrieval show that using DCT intrinsic orientation achieves performance on a par with SIFT, but with only 60% of its features; replacing the SIFT description with DIFT reduces dimensions from 128 to 32 and improves precision. Image reconstruction resulting from DIFT is presented to show another of its advantages over SIFT.