Vision Transformers (ViTs) have revolutionized computer vision, yet their self-attention mechanism lacks explicit spatial inductive biases, leading to suboptimal performance on spatially-structured tasks. Existing approaches introduce data-independent spatial decay based on fixed distance metrics, applying uniform attention weighting regardless of image content and limiting adaptability to diverse visual scenarios. Inspired by recent advances in large language models where content-aware gating mechanisms (e.g., GLA, HGRN2, FOX) significantly outperform static alternatives, we present the first successful adaptation of data-dependent spatial decay to 2D vision transformers. We introduce Spatial Decay Transformer (SDT), featuring a novel Context-Aware Gating (CAG) mechanism that generates dynamic, data-dependent decay for patch interactions. Our approach learns to modulate spatial attention based on both content relevance and spatial proximity. We address the fundamental challenge of 1D-to-2D adaptation through a unified spatial-content fusion framework that integrates manhattan distance-based spatial priors with learned content representations. Extensive experiments on ImageNet-1K classification and generation tasks demonstrate consistent improvements over strong baselines. Our work establishes data-dependent spatial decay as a new paradigm for enhancing spatial attention in vision transformers.

Non-Rigid Structure-from-Motion (NRSfM) is a classic 3D vision problem, where a 2D sequence is taken as input to estimate the corresponding 3D sequence. Recently, the deep neural networks have greatly advanced the task of NRSfM. However, existing deep NRSfM methods still have limitations in handling the inherent sequence property and motion ambiguity associated with the NRSfM problem. In this paper, we revisit deep NRSfM from two perspectives to address the limitations of current deep NRSfM methods : (1) canonicalization and (2) sequence modeling. We propose an easy-to-implement per-sequence canonicalization method as opposed to the previous per-dataset canonicalization approaches. With this in mind, we propose a sequence modeling method that combines temporal information and subspace constraint. As a result, we have achieved a more optimal NRSfM reconstruction pipeline compared to previous efforts. The effectiveness of our method is verified by testing the sequence-to-sequence deep NRSfM pipeline with corresponding regularization modules on several commonly used datasets.

Video-to-audio (V2A) generation utilizes visual-only video features to produce realistic sounds that correspond to the scene. However, current V2A models often lack fine-grained control over the generated audio, especially in terms of loudness variation and the incorporation of multi-modal conditions. To overcome these limitations, we introduce Tri-Ergon, a diffusion-based V2A model that incorporates textual, auditory, and pixel-level visual prompts to enable detailed and semantically rich audio synthesis. Additionally, we introduce Loudness Units relative to Full Scale (LUFS) embedding, which allows for precise manual control of the loudness changes over time for individual audio channels, enabling our model to effectively address the intricate correlation of video and audio in real-world Foley workflows. Tri-Ergon is capable of creating 44.1 kHz high-fidelity stereo audio clips of varying lengths up to 60 seconds, which significantly outperforms existing state-of-the-art V2A methods that typically generate mono audio for a fixed duration.

The Audio-Visual Event Localization (AVEL) task aims to temporally locate and classify video events that are both audible and visible. Most research in this field assumes a closed-set setting, which restricts these models’ ability to handle test data containing event categories absent (unseen) during training. Recently, a few studies have explored AVEL in an open-set setting, enabling the recognition of unseen events as "unknown", but without providing category-specific semantics. In this paper, we advance the field by introducing the Open-Vocabulary Audio-Visual Event Localization (OV-AVEL) problem, which requires localizing audio-visual events and predicting explicit categories for both seen and unseen test data at inference. To address this new task, we propose the OV-AVEBench dataset, comprising 24,800 videos across 67 real-life audiovisual scenes (seen:unseen = 46:21), each with manual segment-level annotation. We also establish three evaluation metrics for this task. Moreover, we investigate two baseline approaches, one training-free and one using a further fine-tuning paradigm. Specifically, we utilize the unified multimodal space from the pretrained ImageBind model to extract audio, visual, and textual (event classes) features. The training-free baseline then determines predictions by comparing the consistency of audio-text and visual-text feature similarities. The fine-tuning baseline incorporates lightweight temporal layers to encode temporal relations within the audio and visual modalities, using OVAVEBench training data for model fine-tuning. We evaluate these baselines on the proposed OV-AVEBench dataset and discuss potential directions for future work in this new field.

Fine-grained video action recognition can be formulated as a video–text matching problem. Previous approaches primarily rely on global video semantics to consolidate video embeddings, often leading to misaligned video–text pairs due to inaccurate atomic-level action understanding. This inaccuracy arises due to i) videos with distinct global semantics may share similar atomic actions or visual appearances, and ii) atomic actions can be momentary, gradual, or not directly aligned with overarching video semantics. Inspired by storyboarding, where a script is segmented into individual shots, we propose a multi-granularity framework, SFAR. SFAR generates fine-grained descriptions of common atomic actions for each global semantic using a large language model. Unlike existing works that refine global semantics with auxiliary video frames, SFAR introduces a filtering metric to ensure correspondence between the descriptions and the global semantics, eliminating the need for direct video involvement and thereby enabling more nuanced recognition of subtle actions. By leveraging both global semantics and fine-grained descriptions, our SFAR effectively identifies prominent frames within videos, thereby improving the accuracy of embedding aggregation. Extensive experiments on various video action recognition datasets demonstrate the competitive performance of our SFAR in supervised, few-shot, and zero-shot settings.

Audio-visual Segmentation (AVS) is conceptualized as a conditional generation task, where audio is considered as the conditional variable for segmenting the sound producer(s). In this case, audio should be extensively explored to maximize its contribution for the final segmentation task. We propose a contrastive conditional latent diffusion model for audio-visual segmentation (AVS) to thoroughly investigate the impact of audio, where the correlation between audio and the final segmentation map is modeled to guarantee the strong correlation between them. To achieve semantic-correlated representation learning, our framework incorporates a latent diffusion model. The diffusion model learns the conditional generation process of the ground-truth segmentation map, resulting in ground-truth aware inference during the denoising process at the test stage. As our model is conditional, it is vital to ensure that the conditional variable contributes to the model output. We thus extensively model the contribution of the audio signal by minimizing the density ratio between the conditional probability of the multimodal data, e.g. conditioned on the audio-visual data, and that of the unimodal data, e.g. conditioned on the audio data only. In this way, our latent diffusion model via density ratio optimization explicitly maximizes the contribution of audio for AVS, which can then be achieved with contrastive learning as a constraint, where the diffusion part serves as the main objective to achieve maximum likelihood estimation, and the density ratio optimization part imposes the constraint. By adopting this latent diffusion model via contrastive learning, we effectively enhance the contribution of audio for AVS. The effectiveness of our solution is validated through experimental results on the benchmark dataset. Code and results are online via our project page: https://github.com/OpenNLPLab/DiffusionAVS.

The aim of audio-visual segmentation (AVS) is to precisely differentiate audible objects within videos down to the pixel level. Traditional approaches often tackle this challenge by combining information from various modalities, where the contribution of each modality is implicitly or explicitly modeled. Nevertheless, the interconnections between different modalities tend to be overlooked in audio-visual modeling. In this paper, inspired by the human ability to mentally simulate the sound of an object and its visual appearance, we introduce a bidirectional generation framework. This framework establishes robust correlations between an object's visual characteristics and its associated sound, thereby enhancing the performance of AVS. To achieve this, we employ a visual-to-audio projection component that reconstructs audio features from object segmentation masks and minimizes reconstruction errors. Moreover, recognizing that many sounds are linked to object movements, we introduce an implicit volumetric motion estimation module to handle temporal dynamics that may be challenging to capture using conventional optical flow methods. To showcase the effectiveness of our approach, we conduct comprehensive experiments and analyses on the widely recognized AVSBench benchmark. As a result, we establish a new state-of-the-art performance level in the AVS benchmark, particularly excelling in the challenging MS3 subset which involves segmenting multiple sound sources. Code is released in: https://github.com/OpenNLPLab/AVS-bidirectional.

Length extrapolation has attracted considerable attention recently since it allows transformers to be tested on longer sequences than those used in training. Previous research has shown that this property can be attained by using carefully designed Relative Positional Encodings (RPEs). While these methods perform well on a variety of corpora, the conditions for length extrapolation have yet to be investigated. This paper attempts to determine what types of RPEs allow for length extrapolation through a thorough mathematical and empirical analysis. We discover that a transformer is certain to possess this property as long as the series that corresponds to the RPE's exponential converges. Two practices are derived from the conditions and examined in language modeling tasks on a variety of corpora. As a bonus from the conditions, we derive a new Theoretical Receptive Field (TRF) to measure the receptive field of RPEs without taking any training steps. Extensive experiments are conducted on the Wikitext-103, Books, Github, and WikiBook datasets to demonstrate the viability of our discovered conditions. We also compare TRF to Empirical Receptive Field (ERF) across different models, showing consistently matched trends on these datasets. Code is released at: https://github.com/OpenNLPLab/Rpe.

The interest in linear complexity models for large language models is on the rise, although their scaling capacity remains uncertain.In this study, we present the scaling laws for linear complexity language models to establish a foundation for their scalability.Specifically, we examine the scaling behaviors of three efficient linear architectures.These include TNL (Qin et al., 2024c), a linear attention model with data-independent decay; HGRN2 (Qin et al., 2024e), a linear RNN with data-dependent decay; and cosFormer2 (Qin et al., 2022b(Qin et al., , 2024a)), a linear attention model without decay.We also include LLaMA as a baseline architecture for comparison with softmax attention.These models were trained with six variants, ranging from 70M to 7B parameters on a 300B-token corpus, and evaluated with a total of 1,376 intermediate checkpoints on various downstream tasks.These tasks include validation loss, commonsense reasoning, and information retrieval and generation.The study reveals that existing linear complexity language models exhibit similar scaling capabilities as conventional transformer-based models while also demonstrating superior linguistic proficiency and knowledge retention.

The fundamental success of large language models hinges upon the efficacious implementation of large-scale distributed training techniques. Nevertheless, building a vast, high-performance cluster featuring high-speed communication interconnectivity is prohibitively costly, and accessible only to prominent entities. In this work, we aim to lower this barrier and democratize large-scale training with limited bandwidth clusters. We propose a new approach called CO2 that introduces local-updating and asynchronous communication to the distributed data-parallel training, thereby facilitating the full overlap of COmmunication with COmputation. CO2 is able to attain a high scalability even on extensive multi-node clusters constrained by very limited communication bandwidth. We further propose the staleness gap penalty and outer momentum clipping techniques together with CO2 to bolster its convergence and training stability. Besides, CO2 exhibits seamless integration with well-established ZeRO-series optimizers which mitigate memory consumption of model states with large model training. We also provide a mathematical proof of convergence, accompanied by the establishment of a stringent upper bound. Furthermore, we validate our findings through an extensive set of practical experiments encompassing a wide range of tasks in the fields of computer vision and natural language processing. These experiments serve to demonstrate the capabilities of CO2 in terms of convergence, generalization, and scalability when deployed across configurations comprising up to 128 A100 GPUs. The outcomes emphasize the outstanding capacity of CO2 to hugely improve scalability, no matter on clusters with 800Gbps RDMA or 80Gbps TCP/IP inter-node connections.

We present Lightning Attention, the first linear attention implementation that maintains a constant training speed for various sequence lengths under fixed memory consumption. Due to the issue with cumulative summation operations (cumsum), previous linear attention implementations cannot achieve their theoretical advantage in a casual setting. However, this issue can be effectively solved by utilizing different attention calculation strategies to compute the different parts of attention. Specifically, we split the attention calculation into intra-blocks and inter-blocks and use conventional attention computation for intra-blocks and linear attention kernel tricks for inter-blocks. This eliminates the need for cumsum in the linear attention calculation. Furthermore, a tiling technique is adopted through both forward and backward procedures to take full advantage of the GPU hardware. To enhance accuracy while preserving efficacy, we introduce TransNormerLLM (TNL), a new architecture that is tailored to our lightning attention. We conduct rigorous testing on standard and self-collected datasets with varying model sizes and sequence lengths. TNL is notably more efficient than other language models. In addition, benchmark results indicate that TNL performs on par with state-of-the-art LLMs utilizing conventional transformer structures. The source code is released at github.com/OpenNLPLab/TransnormerLLM.

We explore the geolocation problem, aiming to localize ground-view images on cartographic maps, without the need of any GPS priors. This task mimics the human wayfinding ability and offers high scalability and robustness by using the compact and semantic representations of maps. Current methods often rely on 2D maps to encode dense contextual information for ground-to-map matching. In this paper, we lift ground-to-map matching to a 2.5D space, where heights of structures (e.g. buildings) provide richer geometric information to guide the matching process. We propose a new approach to learning representative embeddings from multi-modal data. Specifically, we establish a projection relationship between 2D and 2.5D space. The projection is further used to combine multi-modal features from the 2D and 2.5D maps using an effective pixel-to-point fusion method. By encoding crucial geometric cues, our method learns discriminative location embeddings for matching panoramic images and maps. Additionally, we construct the first large-scale multi-modal geolocation dataset to validate our method and facilitate future research. Both single-image based and route based geolocation experiments are conducted to test our method. Extensive experiments demonstrate that the proposed method achieves significantly higher geolocation accuracy and faster convergence than previous 2D map-based approaches.

The Text to Audible-Video Generation (TAVG) task involves generating videos with accompanying audio based on text descriptions. Achieving this requires skillful alignment of both audio and video elements. To support research in this field, we have developed a comprehensive Text to Audible-Video Generation Benchmark (TAVGBench), which contains over 1.7 million clips with a total duration of 11.8 thousand hours. We propose an automatic annotation pipeline to ensure each audible video has detailed descriptions for both its audio and video contents. We also introduce the Audio-Visual Harmoni score (AVHScore) to provide a quantitative measure of the alignment between the generated audio and video modalities. Additionally, we present a baseline model for TAVG called TAVDiffusion, which uses a two-stream latent diffusion model to provide a fundamental starting point for further research in this area. We achieve the alignment of audio and video by employing cross-attention and contrastive learning. Through extensive experiments and evaluations on TAVGBench, we demonstrate the effectiveness of our proposed model under both conventional metrics and our proposed metrics. The dataset and code can be found on this page https://npucvr.github.io/TAVGBench/ and on github https://github.com/OpenNLPLab/TAVGBench.

Various linear complexity models, such as Linear Transformer (LinFormer), State Space Model (SSM), and Linear RNN (LinRNN), have been proposed to replace the conventional softmax attention in Transformer structures. However, the optimal design of these linear models is still an open question. In this work, we attempt to answer this question by finding the best linear approximation to softmax attention from a theoretical perspective. We start by unifying existing linear complexity models as the linear attention form and then identify three conditions for the optimal linear attention design: (1) Dynamic memory ability; (2) Static approximation ability; (3) Least parameter approximation. We find that none of the current linear models meet all three conditions, resulting in suboptimal performance. Instead, we propose Meta Linear Attention (MetaLA) as a solution that satisfies these conditions. Our experiments on Multi-Query Associative Recall (MQAR) task, language modeling, image classification, and Long-Range Arena (LRA) benchmark demonstrate that MetaLA is more effective than the existing linear models.

Directly regressing the non-rigid shape and camera pose from the individual 2D frame is ill-suited to the Non-Rigid Structure-from-Motion (NRSfM) problem. This frame-by-frame 3D reconstruction pipeline overlooks the inherent spatial-temporal nature of NRSfM, i.e., reconstructing the 3D sequence from the input 2D sequence. In this paper, we propose to solve deep sparse NRSfM from a sequence-to-sequence translation perspective, where the input 2D keypoints sequence is taken as a whole to reconstruct the corresponding 3D keypoints sequence in a self-supervised manner. First, we apply a shape-motion predictor on the input sequence to obtain an initial sequence of shapes and corresponding motions. Then, we propose the Context Layer, which enables the deep learning framework to effectively impose overall constraints on sequences based on the structural characteristics of non-rigid sequences. The Context Layer constructs modules for imposing the self-expressiveness regularity on non-rigid sequences with multi-head attention (MHA) as the core, together with the use of temporal encoding, both of which act simultaneously to constitute constraints on non-rigid sequences in the deep framework. Experimental results across different datasets such as Human3.6M, CMU Mocap, and InterHand prove the superiority of our framework. The code will be made publicly available.

We explore a new task for audio-visual-language modeling called fine-grained audible video description (FAVD). It aims to provide detailed textual descriptions for the given audible videos, including the appearance and spatial locations of each object, the actions of moving objects, and the sounds in videos. Existing visual-language modeling tasks often concentrate on visual cues in videos while undervaluing the language and audio modalities. On the other hand, FAVD requires not only audio-visual-language modeling skills but also paragraph-level language generation abilities. We construct the first fine-grained audible video description benchmark (FAVDBench) to facilitate this research. For each video clip, we first provide a one-sentence summary of the video, i.e., the caption, followed by 4–6 sentences describing the visual details and 1–2 audio-related descriptions at the end. The descriptions are provided in both English and Chinese. We create two new metrics for this task: an EntityScore to gauge the completeness of entities in the visual descriptions, and an AudioScore to assess the audio descriptions. As a preliminary approach to this task, we propose an audio-visual-language transformer that extends existing video captioning model with an additional audio branch. We combine the masked language modeling and auto-regressive language modeling losses to optimize our model so that it can produce paragraph-level descriptions. We illustrate the efficiency of our model in audio-visual-language modeling by evaluating it against the proposed benchmark using both conventional captioning metrics and our proposed metrics. We further put our benchmark to the test in video generation models, demonstrating that employing fine-grained video descriptions can create more intricate videos than using captions. Code and dataset are available at https://github.com/OpenNLPLab/FAVDBench. Our online benchmark is available at www.avlbench.opennlplab.cn.

Self-supervised audio-visual source localization aims to locate sound-source objects in video frames without extra annotations. Recent methods often approach this goal with the help of contrastive learning, which assumes only the audio and visual contents from the same video are positive samples for each other. However, this assumption would suffer from false negative samples in real-world training. For example, for an audio sample, treating the frames from the same audio class as negative samples may mislead the model and therefore harm the learned representations (e.g., the audio of a siren wailing may reasonably correspond to the ambulances in multiple images). Based on this observation, we propose a new learning strategy named False Negative Aware Contrastive (FNAC) to mitigate the problem of misleading the training with such false negative samples. Specifically, we utilize the intra-modal similarities to identify potentially similar samples and construct corresponding adjacency matrices to guide contrastive learning. Further, we propose to strengthen the role of true negative samples by explicitly leveraging the visual features of sound sources to facilitate the differentiation of authentic sounding source regions. FNAC achieves state-of-the-art performances on Flickr-SoundNet, VGG-Sound, and AVSBench, which demonstrates the effectiveness of our method in mitigating the false negative issue. The code is available at https://github.com/OpenNLPLab/FNAC_AVL.

github.com/OpenNLPLab/ETSC-Exact-Toeplitz-to

We propose an Explicit Conditional Multimodal Variational Auto-Encoder (ECMVAE) for audio-visual segmentation (AVS), aiming to segment sound sources in the video sequence. Existing AVS methods focus on implicit feature fusion strategies, where models are trained to fit the discrete samples in the dataset. With a limited and less diverse dataset, the resulting performance is usually unsatisfactory. In contrast, we address this problem from an effective representation learning perspective, aiming to model the contribution of each modality explicitly. Specifically, we find that audio contains critical category information of the sound producers, and visual data provides candidate sound producer(s). Their shared information corresponds to the target sound producer(s) shown in the visual data. In this case, cross-modal shared representation learning is especially important for AVS. To achieve this, our ECMVAE factorizes the representations of each modality with a modality-shared representation and a modality-specific representation. An orthogonality constraint is applied between the shared and specific representations to maintain the exclusive attribute of the factorized latent code. Further, a mutual information maximization regularizer is introduced to achieve extensive exploration of each modality. Quantitative and qualitative evaluations on the AVSBench demonstrate the effectiveness of our approach, leading to a new state-of-the-art for AVS, with a 3.84 mIOU performance leap on the challenging MS3 subset for multiple sound source segmentation.

Transformers have surpassed RNNs in popularity due to their superior abilities in parallel training and long-term dependency modeling. Recently, there has been a renewed interest in using linear RNNs for efficient sequence modeling. These linear RNNs often employ gating mechanisms in the output of the linear recurrence layer while ignoring the significance of using forget gates within the recurrence. In this paper, we propose a gated linear RNN model dubbed Hierarchically Gated Recurrent Neural Network (HGRN), which includes forget gates that are lower bounded by a learnable value. The lower bound increases monotonically when moving up layers. This allows the upper layers to model long-term dependencies and the lower layers to model more local, short-term dependencies. Experiments on language modeling, image classification, and long-range arena benchmarks showcase the efficiency and effectiveness of our proposed model. The source code is available at https://github.com/OpenNLPLab/HGRN.

Vision transformers have shown great success on numerous computer vision tasks. However, their central component, softmax attention, prohibits vision transformers from scaling up to high-resolution images, due to both the computational complexity and memory footprint being quadratic. Linear attention was introduced in natural language processing (NLP) which reorders the self-attention mechanism to mitigate a similar issue, but directly applying existing linear attention to vision may not lead to satisfactory results. We investigate this problem and point out that existing linear attention methods ignore an inductive bias in vision tasks, i.e., 2D locality. In this article, we propose Vicinity Attention, which is a type of linear attention that integrates 2D locality. Specifically, for each image patch, we adjust its attention weight based on its 2D Manhattan distance from its neighbouring patches. In this case, we achieve 2D locality in a linear complexity where the neighbouring image patches receive stronger attention than far away patches. In addition, we propose a novel Vicinity Attention Block that is comprised of Feature Reduction Attention (FRA) and Feature Preserving Connection (FPC) in order to address the computational bottleneck of linear attention approaches, including our Vicinity Attention, whose complexity grows quadratically with respect to the feature dimension. The Vicinity Attention Block computes attention in a compressed feature space with an extra skip connection to retrieve the original feature distribution. We experimentally validate that the block further reduces computation without degenerating the accuracy. Finally, to validate the proposed methods, we build a linear vision transformer backbone named Vicinity Vision Transformer (VVT). Targeting general vision tasks, we build VVT in a pyramid structure with progressively reduced sequence length. We perform extensive experiments on CIFAR-100, ImageNet-1 k, and ADE20 K datasets to validate the effectiveness of our method. Our method has a slower growth rate in terms of computational overhead than previous transformer-based and convolution-based networks when the input resolution increases. In particular, our approach achieves state-of-the-art image classification accuracy with 50% fewer parameters than previous approaches.

This work studies the task of glossification, of which the aim is to em transcribe natural spoken language sentences for the Deaf (hard-of-hearing) community to ordered sign language glosses. Previous sequence-to-sequence language models trained with paired sentence-gloss data often fail to capture the rich connections between the two distinct languages, leading to unsatisfactory transcriptions. We observe that despite different grammars, glosses effectively simplify sentences for the ease of deaf communication, while sharing a large portion of vocabulary with sentences. This has motivated us to implement glossification by executing a collection of editing actions, e.g. word addition, deletion, and copying, called editing programs, on their natural spoken language counterparts. Specifically, we design a new neural agent that learns to synthesize and execute editing programs, conditioned on sentence contexts and partial editing results. The agent is trained to imitate minimal editing programs, while exploring more widely the program space via policy gradients to optimize sequence-wise transcription quality. Results show that our approach outperforms previous glossification models by a large margin, improving the BLEU-4 score from 16.45 to 18.89 on RWTH-PHOENIX-WEATHER-2014T and from 18.38 to 21.30 on CSL-Daily.

We propose a new video camouflaged object detection (VCOD) framework that can exploit both short-term dynamics and long-term temporal consistency to detect camouflaged objects from video frames. An essential property of camouflaged objects is that they usually exhibit patterns similar to the background and thus make them hard to identify from still images. Therefore, effectively handling temporal dynamics in videos becomes the key for the VCOD task as the camouflaged objects will be noticeable when they move. However, current VCOD methods often leverage homography or optical flows to represent motions, where the detection error may accumulate from both the motion estimation error and the segmentation error. On the other hand, our method unifies motion estimation and object segmentation within a single optimization framework. Specifically, we build a dense correlation volume to implicitly capture motions between neighbouring frames and utilize the final segmentation supervision to optimize the implicit motion estimation and segmentation jointly. Furthermore, to enforce temporal consistency within a video sequence, we jointly utilize a spatio-temporal transformer to refine the short-term predictions. Extensive experiments on VCOD benchmarks demonstrate the architectural effectiveness of our approach. We also provide a large-scale VCOD dataset named MoCA-Mask with pixel-level handcrafted ground-truth masks and construct a comprehensive VCOD bench-mark with previous methods to facilitate research in this direction. Dataset Link: https://xueliancheng.github.io/SLT-Net-project.

Linear transformers aim to reduce the quadratic space-time complexity of vanilla transformers.However, they usually suffer from degraded performances on various tasks and corpora.In this paper, we examine existing kernel-based linear transformers and identify two key issues that lead to such performance gaps: 1) unbounded gradients in the attention computation adversely impact the convergence of linear transformer models; 2) attention dilution which trivially distributes attention scores over long sequences while neglecting neighbouring structures.To address these issues, we first identify that the scaling of attention matrices is the devil in unbounded gradients, which turns out unnecessary in linear attention as we show theoretically and empirically.To this end, we propose a new linear attention that replaces the scaling operation with a normalization to stabilize gradients.For the issue of attention dilution, we leverage a diagonal attention to confine attention to only neighbouring tokens in early layers.Benefiting from the stable gradients and improved attention, our new linear transformer model, TRANSNORMER, demonstrates superior performance on text classification and language modeling tasks, as well as on the challenging Long-Range Arena benchmark, surpassing vanilla transformer and existing linear variants by a clear margin while being significantly more space-time efficient.The code is available at TRANSNORMER.

Abstract Although deep learning-based methods have dominated stereo matching leaderboards by yielding unprecedented disparity accuracy, their inference time is typically slow, i.e., less than 4 FPS for a pair of 540p images. The main reason is that the leading methods employ time-consuming 3D convolutions applied to a 4D feature volume. A common way to speed up the computation is to downsample the feature volume, but this loses high-frequency details. To overcome these challenges, we propose a displacement-invariant cost computation module to compute the matching costs without needing a 4D feature volume. Rather, costs are computed by applying the same 2D convolution network on each disparity-shifted feature map pair independently. Unlike previous 2D convolution-based methods that simply perform context mapping between inputs and disparity maps, our proposed approach learns to match features between the two images. We also propose an entropy-based refinement strategy to refine the computed disparity map, which further improves the speed by avoiding the need to compute a second disparity map on the right image. Extensive experiments on standard datasets (SceneFlow, KITTI, ETH3D, and Middlebury) demonstrate that our method achieves competitive accuracy with much less inference time. On typical image sizes (e.g., $$540\times 960$$ 540 × 960 ), our method processes over 100 FPS on a desktop GPU, making our method suitable for time-critical applications such as autonomous driving. We also show that our approach generalizes well to unseen datasets, outperforming 4D-volumetric methods. We will release the source code to ensure the reproducibility.

Video deblurring models exploit consecutive frames to remove blurs from camera shakes and object motions. In order to utilize neighboring sharp patches, typical methods rely mainly on homography or optical flows to spatially align neighboring blurry frames. However, such explicit approaches are less effective in the presence of fast motions with large pixel displacements. In this work, we propose a novel implicit method to learn spatial correspondence among blurry frames in the feature space. To construct distant pixel correspondences, our model builds a correlation volume pyramid among all the pixel-pairs between neigh-boring frames. To enhance the features of the reference frame, we design a correlative aggregation module that maximizes the pixel-pair correlations with its neighbors based on the volume pyramid. Finally, we feed the aggregated features into a reconstruction module to obtain the restored frame. We design a generative adversarial paradigm to optimize the model progressively. Our proposed method is evaluated on the widely-adopted DVD dataset, along with a newly collected High-Frame-Rate (1000 fps) Dataset for Video Deblurring (HFR-DVD). Quantitative and qualitative experiments show that our model performs favorably on both datasets against previous state-of-the-art methods, confirming the benefit of modeling all-range spatial correspondence for video deblurring.

Two-view structure-from-motion (SfM) is the cornerstone of 3D reconstruction and visual SLAM. Existing deep learning-based approaches formulate the problem by either recovering absolute pose scales from two consecutive frames or predicting a depth map from a single image, both of which are ill-posed problems. In contrast, we propose to revisit the problem of deep two-view SfM by leveraging the well-posedness of the classic pipeline. Our method consists of 1) an optical flow estimation network that predicts dense correspondences between two frames; 2) a normalized pose estimation module that computes relative camera poses from the 2D optical flow correspondences, and 3) a scale-invariant depth estimation network that leverages epipolar geometry to reduce the search space, refine the dense correspondences, and estimate relative depth maps. Extensive experiments show that our method outperforms all state-of-the-art two-view SfM methods by a clear margin on KITTI depth, KITTI VO, MVS, Scenes11, and SUN3D datasets in both relative pose and depth estimation.

Visual and audio signals often coexist in natural environments, forming audio-visual events (AVEs). Given a video, we aim to localize video segments containing an AVE and identify its category. In order to learn discriminative features for a classifier, it is pivotal to identify the helpful (or positive) audio-visual segment pairs while filtering out the irrelevant ones, regardless whether they are synchronized or not. To this end, we propose a new positive sample propagation (PSP) module to discover and exploit the closely related audio-visual pairs by evaluating the relationship within every possible pair. It can be done by constructing an all-pair similarity map between each audio and visual segment, and only aggregating the features from the pairs with high similarity scores. To encourage the network to extract high correlated features for positive samples, a new audio-visual pair similarity loss is proposed. We also propose a new weighting branch to better exploit the temporal correlations in weakly supervised setting. We perform extensive experiments on the public AVE dataset and achieve new state-of-the-art accuracy in both fully and weakly supervised settings, thus verifying the effectiveness of our method.

Existing RGB-D saliency detection models do not explicitly encourage RGB and depth to achieve effective multi-modal learning. In this paper, we introduce a novel multistage cascaded learning framework via mutual information minimization to explicitly model the multi-modal information between RGB image and depth data. Specifically, we first map the feature of each mode to a lower dimensional feature vector, and adopt mutual information minimization as a regularizer to reduce the redundancy between appearance features from RGB and geometric features from depth. We then perform multi-stage cascaded learning to impose the mutual information minimization constraint at every stage of the network. Extensive experiments on benchmark RGB-D saliency datasets illustrate the effectiveness of our framework. Further, to prosper the development of this field, we contribute the largest (7× larger than NJU2K) COME15K dataset, which contains 15,625 image pairs with high quality polygon-/scribble-/object-/instance-/rank-level annotations. Based on these rich labels, we additionally construct four new benchmarks with strong baselines and observe some interesting phenomena, which can motivate future model design. Source code and dataset are available at https://github.com/JingZhang617/cascaded_rgbd_sod.

Single-image super-resolution (SR) and multi-frame SR are two ways to super resolve low-resolution images. Single-Image SR generally handles each image independently, but ignores the temporal information implied in continuing frames. Multi-frame SR is able to model the temporal dependency via capturing motion information. However, it relies on neighbouring frames which are not always available in the real world. Meanwhile, slight camera shake easily causes heavy motion blur on long-distance-shot low-resolution images. To address these problems, a Blind Motion Deblurring Super-Reslution Networks, BMDSRNet, is proposed to learn dynamic spatio-temporal information from single static motion-blurred images. Motion-blurred images are the accumulation over time during the exposure of cameras, while the proposed BMDSRNet learns the reverse process and uses three-streams to learn Bidirectional spatio-temporal information based on well designed reconstruction loss functions to recover clean high-resolution images. Extensive experiments demonstrate that the proposed BMDSRNet outperforms recent state-of-the-art methods, and has the ability to simultaneously deal with image deblurring and SR.

Existing deep learning methods for image deblurring typically train models using pairs of sharp images and their blurred counterparts. However, synthetically blurring images does not necessarily model the blurring process in real-world scenarios with sufficient accuracy. To address this problem, we propose a new method which combines two GAN models, i.e., a learning-to-Blur GAN (BGAN) and learning-to-DeBlur GAN (DBGAN), in order to learn a better model for image deblurring by primarily learning how to blur images. The first model, BGAN, learns how to blur sharp images with unpaired sharp and blurry image sets, and then guides the second model, DBGAN, to learn how to correctly deblur such images. In order to reduce the discrepancy between real blur and synthesized blur, a relativistic blur loss is leveraged. As an additional contribution, this paper also introduces a Real-World Blurred Image (RWBI) dataset including diverse blurry images. Our experiments show that the proposed method achieves consistently superior quantitative performance as well as higher perceptual quality on both the newly proposed dataset and the public GOPRO dataset.

To reduce the human efforts in neural network design, Neural Architecture Search (NAS) has been applied with remarkable success to various high-level vision tasks such as classification and semantic segmentation. The underlying idea for the NAS algorithm is straightforward, namely, to allow the network the ability to choose among a set of operations (\eg convolution with different filter sizes), one is able to find an optimal architecture that is better adapted to the problem at hand. However, so far the success of NAS has not been enjoyed by low-level geometric vision tasks such as stereo matching. This is partly due to the fact that state-of-the-art deep stereo matching networks, designed by humans, are already sheer in size. Directly applying the NAS to such massive structures is computationally prohibitive based on the currently available mainstream computing resources. In this paper, we propose the first \emph{end-to-end} hierarchical NAS framework for deep stereo matching by incorporating task-specific human knowledge into the neural architecture search framework. Specifically, following the gold standard pipeline for deep stereo matching (\ie, feature extraction -- feature volume construction and dense matching), we optimize the architectures of the entire pipeline jointly. Extensive experiments show that our searched network outperforms all state-of-the-art deep stereo matching architectures and is ranked at the top 1 accuracy on KITTI stereo 2012, 2015, and Middlebury benchmarks, as well as the top 1 on SceneFlow dataset with a substantial improvement on the size of the network and the speed of inference. Code available at https://github.com/XuelianCheng/LEAStereo.

Learning matching costs has been shown to be critical to the success of the state-of-the-art deep stereo matching methods, in which 3D convolutions are applied on a 4D feature volume to learn a 3D cost volume. However, this mechanism has never been employed for the optical flow task. This is mainly due to the significantly increased search dimension in the case of optical flow computation, \ie, a straightforward extension would require dense 4D convolutions in order to process a 5D feature volume, which is computationally prohibitive. This paper proposes a novel solution that is able to bypass the requirement of building a 5D feature volume while still allowing the network to learn suitable matching costs from data. Our key innovation is to decouple the connection between 2D displacements and learn the matching costs at each 2D displacement hypothesis independently, \ie, displacement-invariant cost learning. Specifically, we apply the same 2D convolution-based matching net independently on each 2D displacement hypothesis to learn a 4D cost volume. Moreover, we propose a displacement-aware projection layer to scale the learned cost volume, which reconsiders the correlation between different displacement candidates and mitigates the multi-modal problem in the learned cost volume. The cost volume is then projected to optical flow estimation through a 2D soft-argmin layer. Extensive experiments show that our approach achieves state-of-the-art accuracy on various datasets, and outperforms all published optical flow methods on the Sintel benchmark. The code is available at https://github.com/jytime/DICL-Flow.

In this paper, we present LidarStereoNet, the first unsupervised Lidar-stereo fusion network, which can be trained in an end-to-end manner without the need of ground truth depth maps. By introducing a novel ``Feedback Loop'' to connect the network input with output, LidarStereoNet could tackle both noisy Lidar points and misalignment between sensors that have been ignored in existing Lidar-stereo fusion work. Besides, we propose to incorporate the piecewise planar model into the network learning to further constrain depths to conform to the underlying 3D geometry. Extensive quantitative and qualitative evaluations on both real and synthetic datasets demonstrate the superiority of our method, which outperforms state-of-the-art stereo matching, depth completion and Lidar-Stereo fusion approaches significantly.

Unsupervised deep learning for optical flow computation has achieved promising results. Most existing deep-net based methods rely on image brightness consistency and local smoothness constraint to train the networks. Their performance degrades at regions where repetitive textures or occlusions occur. In this paper, we propose Deep Epipolar Flow, an unsupervised optical flow method which incorporates global geometric constraints into network learning. In particular, we investigate multiple ways of enforcing the epipolar constraint in flow estimation. To alleviate a ``chicken-and-egg'' type of problem encountered in dynamic scenes where multiple motions may be present, we propose a low-rank constraint as well as a union-of-subspaces constraint for training. Experimental results on various benchmarking datasets show that our method achieves competitive performance compared with supervised methods and outperforms state-of-the-art unsupervised deep-learning methods.

Camera shake or target movement often leads to undesired blur effects in videos captured by a hand-held camera. Despite significant efforts having been devoted to video-deblur research, two major challenges remain: 1) how to model the spatio-temporal characteristics across both the spatial domain (i.e., image plane) and the temporal domain (i.e., neighboring frames) and 2) how to restore sharp image details with respect to the conventionally adopted metric of pixel-wise errors. In this paper, to address the first challenge, we propose a deblurring network (DBLRNet) for spatial-temporal learning by applying a 3D convolution to both the spatial and temporal domains. Our DBLRNet is able to capture jointly spatial and temporal information encoded in neighboring frames, which directly contributes to the improved video deblur performance. To tackle the second challenge, we leverage the developed DBLRNet as a generator in the generative adversarial network (GAN) architecture and employ a content loss in addition to an adversarial loss for efficient adversarial training. The developed network, which we name as deblurring GAN, is tested on two standard benchmarks and achieves the state-of-the-art performance.

This paper is concerned with the problem of how to better exploit 3D geometric information for dense semantic image labeling. Existing methods often treat the available 3D geometry information (e.g., 3D depth-map) simply as an additional image channel besides the R-G-B color channels, and apply the same technique for RGB image labeling. In this paper, we demonstrate that directly performing 3D convolution in the framework of a residual connected 3D voxel top-down modulation network can lead to superior results. Specifically, we propose a 3D semantic labeling method to label outdoor street scenes whenever a dense depth map is available. Experiments on the “Synthia” and “Cityscape” datasets show our method outperforms the state-of-the-art methods, suggesting such a simple 3D representation is effective in incorporating 3D geometric information.

Feature tracking is a fundamental problem in computer vision, with applications in many computer vision tasks, such as visual SLAM and action recognition. This paper introduces a novel multi-body feature tracker that exploits a multi-body rigidity assumption to improve tracking robustness under a general perspective camera model. A conventional approach to addressing this problem would consist of alternating between solving two subtasks: motion segmentation and feature tracking under rigidity constraints for each segment. This approach, however, requires knowing the number of motions, as well as assigning points to motion groups, which is typically sensitive to the motion estimates. By contrast, here, we introduce a segmentationfree solution to multi-body feature tracking that bypasses the motion assignment step and reduces to solving a series of subproblems with closed-form solutions. Our experiments demonstrate the benefits of our approach in terms of tracking accuracy and robustness to noise.

The problems of motion segmentation and face clustering can be addressed in a framework of subspace clustering methods. In this paper, we tackle the more general problem of clustering data points lying in a union of low-dimensional linear(or affine) subspaces, which can be naturally applied in motion segmentation and face clustering. For data points drawn from linear (or affine) subspaces, we propose a novel algorithm called Null Space Clustering (NSC), utilizing the null space of the data matrix to construct the affinity matrix. To better deal with noise and outliers, it is converted to an equivalent problem with Frobenius norm minimization, which can be solved efficiently. We demonstrate that the proposed NSC leads to improved performance in terms of clustering accuracy and efficiency when compared to state-of-the-art algorithms on two well-known datasets, i.e., Hopkins 155 and Extended Yale B.