CLIP is a seminal multimodal model that maps images and text into a shared representation space by contrastive learning on billions of image–caption pairs. Inspired by the rapid progress of large language models (LLMs), we investigate how the superior linguistic understanding and broad world knowledge of LLMs can further strengthen CLIP—particularly in handling long, complex captions. We introduce an efficient fine-tuning framework that embeds an LLM into a pretrained CLIP while incurring almost the same training cost as regular CLIP fine-tuning. Our method first “embedding-izes” the LLM for the CLIP setting, then couples it to the pretrained CLIP vision encoder through a lightweight adaptor trained on only a few million image–caption pairs. With this strategy we achieve large performance gains—without large-scale retraining—over state-of-the-art CLIP variants such as EVA02 and SigLIP-2. The LLM-enhanced CLIP delivers consistent improvements across a wide spectrum of downstream tasks, including linear-probe classification, zero-shot image–text retrieval with both short and long captions (in English and other languages), zero-shot/supervised image segmentation, object detection, and used as tokenizer for multimodal large-model benchmarks.

Recent diffusion-based models have significantly improved inpainting quality. However, existing methods struggle with multi-task inpainting due to conflicting optimization objectives, and current datasets are typically limited to task-specific scenarios, hindering joint training. To address these challenges, we propose MagicPaint, a unified diffusion-based inpainting model that supports object addition, removal, and unconditional inpainting across both text and image modalities. MagicPaint semantically decouples operation types and target content by learnable tokens in MMToken Module, effectively reconciling conflicting optimization objectives and enabling robust multi-task, multi-modal inpainting. Besides, a novel inpainting paradigm named MagicMask, encodes operating intent directly into the mask and applies a mask loss for spatially precise supervision. In addition, existing inpainting datasets are insufficient for multi-task and multi-modal scenarios, limiting the capability of inpainting models. Thus, we further introduce a new dataset comprising 2.1M image tuples. It is dedicatedly designed to support diverse inpainting scenarios and significantly improves upon existing datasets, particularly in object removal. Through efforts from both model and data perspectives, MagicPaint enables users to operate anything—add, remove or inpaint content which is specified through either text or image modalities in a seamless and unified manner. Extensive experiments demonstrate that MagicPaint achieves state-of-the-art performance across three key tasks (i.e., text-guided addition, image-guided addition, and object removal) and produces outputs with superior visual consistency and contextual fidelity compared to existing methods.

Recent models like OpenAI’s O1 and DeepSeek’s R1, which utilize test-time scaling techniques, have demonstrated remarkable improvements in reasoning capabilities. We anticipate that in the near future, multimodal models will also experience significant breakthroughs in multimodal reasoning. This will require some highly challenging and specialized evaluations. As one of the most crucial real-world applications of multimodal models, visual agents require complex and comprehensive capabilities such as spatial planning and vision-in-the-chain type reasoning. These capabilities are currently lacking in existing multimodal benchmarks. In this paper, we introduce MageBench, a Multimodal reasoning benchmark built upon light-weight AGEnt environments that pose significant reasoning challenges and hold substantial practical value. The results show that only a few product-level models are better than random acting, and all of them are far inferior to human level. We analyze and summarize their errors and capability gaps in visual planning. Furthermore, we found that rule-based RL can significantly boost visual reasoning capabilities. This highlights that our benchmark could serve as a valuable testing ground for the emerging field of agentic RL research.

Video-based large language models (Video-LLMs) have been recently introduced, targeting both fundamental improvements in perception and comprehension, and a diverse range of user inquiries. In pursuit of the ultimate goal of achieving artificial general intelligence, a truly intelligent Video-LLM model should not only see and understand the surroundings, but also possess human-level commonsense, and make well-informed decisions for users. To guide the development of such a model, the establishment of a robust and comprehensive evaluation system becomes crucial. To this end, this paper proposes Video-Bench, a new comprehensive benchmark along with a toolkit specifically designed for evaluating Video-LLMs. The benchmark comprises 10 meticulously crafted tasks, evaluating the capabilities of Video-LLMs across three distinct levels: video-exclusive understanding, prior knowledge-based question-answering, and comprehension and decision-making. In addition, we introduce an automatic toolkit tailored to process model outputs for various tasks, facilitating the calculation of metrics and conveniently generating final scores. We evaluate 9 representative Video-LLMs using Video-Bench. The findings reveal that current Video-LLMs still fall considerably short of achieving human-like comprehension and analysis of real-world video, and offer valuable insights for future research directions. The benchmark and toolkit are available at https://github.com/PKU-YuanGroup/Video-Bench.

Abstract The limited availability of 3D medical image datasets, due to privacy concerns and high collection or annotation costs, poses significant challenges in the field of medical imaging. There are few solutions for realistic 3D medical image synthesis due to difficulties in backbone design and fewer 3D training samples compared to 2D counterparts. In this paper, we propose GEM-3D , a novel generative approach to the synthesis of 3D medical images and the enhancement of existing datasets using conditional diffusion models. Our method begins with a 2D slice, noted as the informed slice to serve the patient prior, and propagates the generation process using a 3D segmentation mask. By decomposing the 3D medical images into editable masks and patient prior information, GEM-3D offers a flexible yet effective solution for generating versatile 3D images from existing datasets. Moreover, as the informed slice contains patient-wise information, GEM-3D can also facilitate counterfactual image synthesis and dataset-level de-enhancement with desired control. Experiments on brain MRI and abdomen CT images demonstrate that GEM-3D is capable of synthesizing high-quality 3D medical images with volumetric consistency, offering a straightforward solution for dataset enhancement during inference. The code is available at https://github.com/HKU-MedAI/GEM-3D .

Hair editing is a long-standing problem in computer vision that demands both fine-grained local control and intuitive user interactions across diverse modalities. Despite the remarkable progress of GANs and diffusion models, existing methods still lack a unified framework that simultaneously supports arbitrary interaction modes (e.g., text, sketch, mask, and reference image) while ensuring precise editing and faithful preservation of irrelevant attributes. In this work, we introduce a novel paradigm that reformulates hair editing as proxy-based hair transfer. Specifically, we leverage the dense and semantically disentangled latent space of StyleGAN for precise manipulation and exploit its feature space for disentangled attribute preservation, thereby decoupling the objectives of editing and preservation. Our framework unifies different modalities by converting editing conditions into distinct transfer proxies, whose features are seamlessly blended to achieve global or local edits. Beyond 2D, we extend our paradigm to 3D-aware settings by incorporating EG3D and PanoHead, where we propose a multi-view boosted hair feature localization strategy together with 3D-tailored proxy generation methods that exploit the inherent properties of 3D-aware generative models. Extensive experiments demonstrate that our method consistently outperforms prior approaches in editing effects, attribute preservation, visual naturalness, and multi-view consistency, while offering unprecedented support for multimodal and mixed-modal interactions.

Representing the cutting-edge technique of text-to-image models, the latest Multimodal Diffusion Transformer (MMDiT) largely mitigates many generation issues existing in previous models. However, we discover that it still suffers from subject neglect or mixing when the input text prompt contains multiple subjects of similar semantics or appearance. We identify three possible ambiguities within the MMDiT architecture that cause this problem: Inter-block Ambiguity, Text Encoder Ambiguity, and Semantic Ambiguity. To address these issues, we propose to repair the ambiguous latent on-the-fly by test-time optimization at early denoising steps. In detail, we design three loss functions: Block Alignment Loss, Text Encoder Alignment Loss, and Overlap Loss, each tailored to mitigate these ambiguities. Despite significant improvements, we observe that semantic ambiguity persists when generating multiple similar subjects, as the guidance provided by overlap loss is not explicit enough. Therefore, we further propose Overlap Online Detection and Back-to-Start Sampling Strategy to alleviate the problem. Experimental results on a newly constructed challenging dataset of similar subjects validate the effectiveness of our approach, showing superior generation quality and much higher success rates over existing methods. The consistent and substantial improvements observed across multiple MMDiT based text-to-image models such as SD3, SD3.5 and FLUX provide strong evidence of the general applicability of our method.

Automatic furniture layout is long desired for convenient interior design. Leveraging the remarkable visual reasoning capabilities of multimodal large language models (MLLMs), recent methods address layout generation in a static manner, lacking the feedback-driven refinement essential for interactive user engagement. We introduce Chat2Layout, a novel interactive furniture layout generation system that extends the functionality of MLLMs into the realm of interactive layout design. To achieve this, we establish a unified vision-question paradigm for in-context learning, enabling seamless communication with MLLMs to steer their behavior without altering model weights. Within this framework, we present a novel training-free visual prompting mechanism. This involves a visual-text prompting technique that assist MLLMs in reasoning about plausible layout plans, followed by an Offline-to-Online search (O2O-Search) method, which identifies the minimal set of informative references to provide exemplars for visual-text prompting. By employing an agent system with MLLMs as the core controller, we enable bidirectional interaction. The agent not only comprehends the 3D environment and user requirements through linguistic and visual perception but also plans tasks and reasons about actions to generate and arrange furniture within the virtual space. Furthermore, the agent iteratively updates based on visual feedback from execution results. Experimental results demonstrate that our approach facilitates language-interactive generation and arrangement for diverse and complex 3D furniture.

Object removal has so far been dominated by the "mask-and-inpaint" paradigm, where the masked region is excluded from the input, leaving models relying on unmasked areas to inpaint the missing region. However, this approach lacks contextual information for the masked area, often resulting in unstable performance. In this work, we introduce SmartEraser, built with a new "removing" paradigm called Masked-Region Guidance. This paradigm retains the masked region in the input, using it as guidance for the removal process. It offers several distinct advantages: (a) it guides the model to accurately identify the object to be removed, preventing its regeneration in the output; (b) since the user mask often extends beyond the object itself, it aids in preserving the surrounding context in the final result. Leveraging this new paradigm, we present Syn4Removal, a large-scale object removal dataset, where instance segmentation data is used to copy and paste objects onto images as removal targets, with the original images serving as ground truths. Experimental results demonstrate that SmartEraser significantly outperforms existing methods, achieving superior performance in object removal, especially in complex scenes with intricate compositions.

We introduce Olympus, a new approach that transforms Multimodal Large Language Models (MLLMs) into a unified framework capable of handling a wide array of computer vision tasks. Utilizing a controller MLLM, Olympus delegates over 20 specialized tasks across images, videos, and 3D objects to dedicated modules. This instruction-based routing enables complex workflows through chained actions without the need for training heavy generative models. Olympus easily integrates with existing MLLMs, expanding their capabilities with comparable performance. Experimental results demonstrate that Olympus achieves an average routing accuracy of 94.75% across 20 tasks and precision of 91.82% in chained action scenarios, showcasing its effectiveness as a universal task router that can solve a diverse range of computer vision tasks.

Recent advancements in in-context segmentation generalists have demonstrated significant success in performing various image segmentation tasks using a limited number of labeled example images. However, real-world applications present challenges due to the variability of support examples, which often exhibit quality issues resulting from various sources and inaccurate labeling. How to extract more robust representations from these examples has always been one of the goals of in-context visual learning. In response, we propose UNICL-SAM, to better model the example distribution and extract robust representations to help in-context segmentation. We incorporate an uncertainty probabilistic module to quantify each example’s reliability during both the training and testing phases. Utilizing this uncertainty estimation, we introduce an uncertainty-guided graph augmentation and feature refinement strategy, aimed at mitigating the impact of high-uncertainty regions to enhance the learning of robust representations. Subsequently, we construct prototypes for each example by aggregating part information, thereby creating reliable in-context instruction that effectively represents fine-grained local semantics. This approach serves as a valuable complement to traditional global pooling features. Experimental results demonstrate the effectiveness of the proposed framework, underscoring its potential for real-world applications.

Rui Wang, Bohao Li, Xiyang Dai, Jianwei Yang, Yi-Ling Chen, Zhen Xing, Yifan Yang, Dongdong Chen, Xipeng Qiu, Zuxuan Wu, Yu-Gang Jiang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

The integration of Rotary Position Embedding (RoPE) in Multimodal Diffusion Transformer (MMDiT) has significantly enhanced text-to-image generation quality. However, the fundamental reliance of self-attention layers on positional embedding versus query-key similarity during generation remains an intriguing question. We present the first mechanistic analysis of RoPE-based MMDiT models (e.g., FLUX), introducing an automated probing strategy that disentangles positional information versus content dependencies by strategically manipulating RoPE during generation. Our analysis reveals distinct dependency patterns that do not straightforwardly correlate with depth, offering new insights into the layer-specific roles in RoPE-based MMDiT. Based on these findings, we propose a training-free, task-specific image editing framework that categorizes editing tasks into three types: position-dependent editing (e.g., object addition), content similarity-dependent editing (e.g., non-rigid editing), and region-preserved editing (e.g., background replacement). For each type, we design tailored key-value injection strategies based on the characteristics of the editing task. Extensive qualitative and quantitative evaluations demonstrate that our method outperforms state-of-the-art approaches, particularly in preserving original semantic content and achieving seamless modifications.

We present I2V3D, a novel framework for animating static images into dynamic videos with precise 3D control, leveraging the strengths of both 3D geometry guidance and advanced generative models. Our approach combines the precision of a computer graphics pipeline, enabling accurate control over elements such as camera movement, object rotation, and character animation, with the visual fidelity of generative AI to produce high-quality videos from coarsely rendered inputs. To support animations with any initial start point and extended sequences, we adopt a two-stage generation process guided by 3D geometry: 1) 3D-Guided Keyframe Generation, where a customized image diffusion model refines rendered keyframes to ensure consistency and quality, and 2) 3D-Guided Video Interpolation, a training-free approach that generates smooth, high-quality video frames between keyframes using bidirectional guidance. Experimental results highlight the effectiveness of our framework in producing controllable, high-quality animations from single input images by harmonizing 3D geometry with generative models. The code for our framework will be publicly released.

In this paper, we empirically demonstrate that natural images can be reconstructed with high fidelity from compressed representations using a simple first-order norm-plus-linear autoregressive (FINOLA) process—without relying on explicit positional information. Through systematic analysis, we observe that the learned coefficient matrices ($\mathbf{A}$ and $\mathbf{B}$) in FINOLA are typically invertible, and their product, $\mathbf{AB}^{-1}$, is diagonalizable across training runs. This structure enables a striking interpretation: FINOLA’s latent dynamics resemble a system of one-way wave equations evolving in a compressed latent space. Under this framework, each image corresponds to a unique solution of these equations. This offers a new perspective on image invariance, suggesting that the underlying structure of images may be governed by simple, invariant dynamic laws. Our findings shed light on a novel avenue for understanding and modeling visual data through the lens of latent-space dynamics and wave propagation.

Abstract Large multimodal language models (MLLMs) such as GPT-4V and GPT-4o have achieved remarkable advancements in understanding and generating multimodal content, showcasing superior quality and capabilities across diverse tasks. However, their deployment faces significant challenges, including slow inference, high computational cost, and impracticality for on-device applications. In contrast, the emergence of small MLLMs, exemplified by the LLava-series models and Phi-3-Vision, offers promising alternatives with faster inference, reduced deployment costs, and the ability to handle domain-specific scenarios. Despite their growing presence, the capability boundaries between large and small MLLMs remain underexplored. In this work, we conduct a systematic and comprehensive evaluation to benchmark both small and large MLLMs, spanning general capabilities such as object recognition, temporal reasoning, and multimodal comprehension, as well as real-world applications in domains like industry and automotive. Our evaluation reveals that small MLLMs can achieve comparable performance to large models in specific scenarios but lag significantly in complex tasks requiring deeper reasoning or nuanced understanding. Furthermore, we identify common failure cases in both small and large MLLMs, highlighting domains where even state-of-the-art models struggle. We hope our findings will guide the research community in pushing the quality boundaries of MLLMs, advancing their usability and effectiveness across diverse applications.

We present SinDiffusion, leveraging denoising diffusion models to capture internal distribution of patches from a single natural image. The default approach of previous GAN-based methods on this problem is to train multiple models at progressive growing scales, which leads to the accumulation of errors and causes characteristic artifacts in generated results. In this paper, we uncover that multiple models at progressive growing scales are not essential for learning from a single image and propose SinDiffusion, a single diffusion-based model trained on a single scale, which is better-suited for this task. Furthermore, we identify that a patch-level receptive field is crucial and effective for diffusion models to capture the image's patch statistics, therefore we redesign an patch-wise denoising network for SinDiffusion. Coupling these two designs enables SinDiffusion to generate more photorealistic and diverse images from a single image compared with GAN-based approaches. SinDiffusion can also be applied to various applications, i.e., text-guided image generation, and image outpainting beyond the capability of SinGAN. Extensive experiments on a wide range of images demonstrate the superiority of SinDiffusion for modeling the patch distribution.

Visual Object Tracking (VOT) aims to estimate the positions of target objects in a video sequence, which is an important vision task with various real-world applications. Depending on whether the initial states of target objects are specified by provided annotations in the first frame or the categories, VOT could be classified as instance tracking (e.g., SOT and VOS) and category tracking (e.g., MOT, MOTS, and VIS) tasks. Different definitions have led to divergent solutions for these two types of tasks, resulting in redundant training expenses and parameter overhead. In this paper, combing the advantages of the best practices developed in both communities, we propose a novel tracking-with-detection paradigm, where tracking supplements appearance priors for detection and detection provides tracking with candidate bounding boxes for the association. Equipped with such a design, a unified tracking model, OmniTracker, is further presented to resolve all the tracking tasks with a fully shared network architecture, model weights, and inference pipeline, eliminating the need for task-specific architectures and reducing redundancy in model parameters. We conduct extensive experimentation on seven prominent tracking datasets of different tracking tasks, including LaSOT, TrackingNet, DAVIS16-17, MOT17, MOTS20, and YTVIS19, and demonstrate that OmniTracker achieves on-par or even better results than both task-specific and unified tracking models.

We introduce pluralistic salient object detection (PSOD), a novel task aimed at generating multiple plausible salient segmentation results for a given input image. Unlike conventional SOD methods that produce a single segmentation mask for salient objects, this new setting recognizes the inherent complexity of real-world images, comprising multiple objects, and the ambiguity in defining salient objects due to different user intentions. To study this task, we present two new SOD datasets "DUTS-MM" and "DUTS-MQ", along with newly designed evaluation metrics. DUTS-MM builds upon the DUTS dataset but enriches the ground-truth mask annotations from three aspects which 1) improves the mask quality especially for boundary and fine-grained structures; 2) alleviates the annotation inconsistency issue; and 3) provides multiple ground-truth masks for images with saliency ambiguity. DUTS-MQ consists of approximately 100K image-mask pairs with human-annotated preference scores, enabling the learning of real human preferences in measuring mask quality. Building upon these two datasets, we propose a simple yet effective pluralistic SOD baseline based on a Mixture-of-Experts (MOE) design. Equipped with two prediction heads, it simultaneously predicts multiple masks using different query prompts and predicts human preference scores for each mask candidate. Extensive experiments and analyses underscore the significance of our proposed datasets and affirm the effectiveness of our PSOD framework.

The highly realistic avatar in the metaverse may lead to deepfakes of facial identity. Malicious users can more easily obtain the three-dimensional structure of faces, thus using deepfake technology to create counterfeit videos with higher realism. To automatically discern facial videos forged with the advancing generation techniques, deepfake detectors need to achieve stronger generalization abilities. Inspired by transfer learning, neural networks pre-trained on other large-scale face-related tasks would provide fundamental features for deepfake detection. We propose a video-level deepfake detection method based on a temporal transformer with a self-supervised audio–visual contrastive learning approach for pre-training the deepfake detector. The proposed method learns motion representations in the mouth region by encouraging the paired video and audio representations to be close while unpaired ones to be diverse. The deepfake detector adopts the pre-trained weights and partially fine-tunes on deepfake datasets. Extensive experiments show that our self-supervised pre-training method can effectively improve the accuracy and robustness of our deepfake detection model without extra human efforts. Compared with existing deepfake detection methods, our proposed method achieves better generalization ability in cross-dataset evaluations.

The rapid advancement of large language models (LLMs) has accelerated the emergence of in-context learning (ICL) as a cutting-edge approach in the natural language processing domain. Recently, ICL has been employed in visual understanding tasks, such as semantic segmentation and image captioning, yielding promising results. However, existing visual ICL framework can not enable producing content across multiple modalities, whicd limits their potential usage scenarios. To address this issue, we present a new ICLframeworkfor visual understanding with multi-modal output enabled. First, we quantize and embed both text and visual prompt into a unified representational space, structured as interleaved in-context sequences. Then a decoder-only sparse transformer architecture is employed to perform generative modeling on them, facilitating in-context learning. Thanks to this design, the model is capable of handling in-context vision understanding tasks with multimodal output in a unified pipeline. Experimental re-sults demonstrate that our model achieves competitive performance compared with specialized models and previous ICL baselines. Overall, our research takes a further step toward unified multimodal in-context learning.

The core of video understanding tasks, such as recognition, captioning, and tracking, is to automatically de-tect objects or actions in a video and analyze their temporal evolution. Despite sharing a common goal, different tasks often rely on distinct model architectures and annotation formats. In contrast, natural language processing benefits from a unified output space, i.e., text sequences, which simplifies the training of powerful foundational language models, such as GPT-3, with extensive training cor-pora. Inspired by this, we seek to unify the output space of video understanding tasks by using languages as labels and additionally introducing time and box tokens. In this way, a variety of video tasks could be formulated as video-grounded token generation. This enables us to address var-ious types of video tasks, including classification (such as action recognition), captioning (covering clip captioning, video question answering, and dense video captioning), and localization tasks (such as visual object tracking) within a fully shared encoder-decoder architecture, following a generative framework. Through comprehensive experiments, we demonstrate such a simple and straightforward idea is quite effective and can achieve state-of-the-art or compet-itive results on seven video benchmarks, providing a novel perspective for more universal video understanding. Code is available at https://github.com/wangjk666/OmniVid.

Face manipulation has ignited the interests of both academia and industry in very recent years. Existing face manipulation methods can be roughly categorized into two types: face attribute editing and face swapping. In this paper, we focus on swapping the identity. But unlike face swapping which only changes the face region, we attempt at a more challenging task: attribute-aware head swapping. Given a source video and a target video, we replace the whole target head with the whole source head while keeping the original target attributes. To address the inherent appearance gap (e.g., hairstyle, face shape), accompanying background incompatibility and lighting difference, our method consists of three key components: 1) a generative rendering-to-real-head model for source head modeling and attribute transfer; 2) a background modeling network to fix the background incompatibility during head swapping; 3) a deep harmonization network to fix remaining issues and makes the final composited result more realistic. We compare our approach to different face manipulation methods and the experimental results demonstrate its superiority for a lot of challenging cases.

ACM Multimedia 2024, Melbourne, Australia, Oct 28 - Nov 1, 2024

Transformer based methods have achieved great success in image inpainting recently. However, we find that these solutions regard each pixel as a token, thus suffering from an information loss issue from two aspects: 1) They downsample the input image into much lower resolutions for efficiency consideration. 2) They quantize 2563RGB values to a small number (such as 512) of quantized color values. The indices of quantized pixels are used as tokens for the inputs and prediction targets of the transformer. To mitigate these issues, we propose a new transformer based framework called “PUT”. Specifically, to avoid input downsampling while maintaining computation efficiency, we design a patch-based auto-encoder P-VQVAE. The encoder converts the masked image into non-overlapped patch tokens and the decoder recovers the masked regions from the inpainted tokens while keeping the unmasked regions unchanged. To eliminate the information loss caused by input quantization, an Un-quantized Transformer is applied. It directly takes features from the P-VQVAE encoder as input without any quantization and only regards the quantized tokens as prediction targets.Furthermore, to make the inpainting process more controllable, we introduce semantic and structural conditions as extra guidance. Extensive experiments show that our method greatly outperforms existing transformer based methods on image fidelity and achieves much higher diversity and better fidelity than state-of-the-art pluralistic inpainting methods on complex large-scale datasets (e.g., ImageNet). Codes are available athttps://github.com/liuqk3/PUT.

Image completion has made tremendous progress with convolutional neural networks (CNNs), because of their powerful texture modeling capacity. However, due to some inherent properties (e.g., local inductive prior, spatial-invariant kernels), CNNs do not perform well in understanding global structures or naturally support pluralistic completion. Recently, transformers demonstrate their power in modeling the long-term relationship and generating diverse results, but their computation complexity is quadratic to input length, thus hampering the application in processing high-resolution images. This paper brings the best of both worlds to pluralistic image completion: appearance prior reconstruction with transformer and texture replenishment with CNN. The former transformer recovers pluralistic coherent structures together with some coarse textures, while the latter CNN enhances the local texture details of coarse priors guided by the high-resolution masked images. To decode diversified outputs from transformers, auto-regressive sampling is the most common method, but with extremely low efficiency. We further overcome this issue by proposing a new decoding strategy, temperature annealing probabilistic sampling (TAPS), which firstly achieves more than 70× speedup of inference at most, meanwhile maintaining the high quality and diversity of the sampled global structures. Moreover, we find the full CNN architecture will lead to suboptimal solutions for guided upsampling. To render more realistic and coherent contents, we design a novel module, named texture-aware guided attention, to concurrently consider the procedures of texture copy and generation, meanwhile raising several important modifications to solve the boundary artifacts. Through dense experiments, we found the proposed method vastly outperforms state-of-the-art methods in terms of four aspects: 1) large performance boost on image fidelity even compared to deterministic completion methods; 2) better diversity and higher fidelity for pluralistic completion; 3) exceptional generalization ability on large masks and generic dataset, like ImageNet. 4) Much higher decoding efficiency over previous auto-regressive based methods.

Image matting is a fundamental and challenging problem in computer vision and graphics. Most existing matting methods leverage a user-supplied trimap as an auxiliary input to produce good alpha matte. However, obtaining high-quality trimap itself is arduous. Recently, some hint-free methods have emerged, however, the matting quality is still far behind the trimap-based methods. The main reason is that, some hints for removing semantic ambiguity and improving matting quality are essential. Apparently, there is a trade-off between interaction cost and matting quality. To balance performance and user-friendliness, we propose an improved deep image matting framework which is trimap-free and only needs sparse user click or scribble interaction to minimize the needed auxiliary constraints while still allowing interactivity. Moreover, we introduce uncertainty estimation that predicts which parts need polishing and conduct uncertainty-guided refinement. To trade off runtime against refinement quality, users can also choose different refinement modes. Experimental results show that our method performs better than existing trimap-free methods and comparably to state-of-the-art trimap-based methods with minimal user effort. Finally, we demonstrate the extensibility of our framework to video human matting without any structure modification, by adding optical flow-based sparse hint propagation and temporal consistency regularization imposed on the single frame.

The intellectual property of deep networks can be easily "stolen" by surrogate model attack. There has been significant progress in protecting the model IP in classification tasks. However, little attention has been devoted to the protection of image processing models. By utilizing consistent invisible spatial watermarks, the work (Zhang et al. 2020) first considered model watermarking for deep image processing networks and demonstrated its efficacy in many downstream tasks. Its success depends on the hypothesis that if a consistent watermark exists in all prediction outputs, that watermark will be learned into the attacker's surrogate model. However, when the attacker uses common data augmentation attacks (e.g., rotate, crop, and resize) during surrogate model training, it will fail because the underlying watermark consistency is destroyed. To mitigate this issue, we propose a new watermarking methodology, "structure consistency", based on which a new deep structure-aligned model watermarking algorithm is designed. Specifically, the embedded watermarks are designed to be aligned with physically consistent image structures, such as edges or semantic regions. Experiments demonstrate that our method is more robust than the baseline in resisting data augmentation attacks. Besides that, we test the generalization ability and robustness of our method to a broader range of adaptive attacks.

Notwithstanding the prominent performance shown in various applications, point cloud recognition models have often suffered from natural corruptions and adversarial perturbations. In this paper, we delve into boosting the general robustness of point cloud recognition, proposing Point-Cloud Contrastive Adversarial Training (PointCAT). The main intuition of PointCAT is encouraging the target recognition model to narrow the decision gap between clean point clouds and corrupted point clouds by devising feature-level constraints rather than logit-level constraints. Specifically, we leverage a supervised contrastive loss to facilitate the alignment and the uniformity of hypersphere representations, and design a pair of centralizing losses with dynamic prototype guidance to prevent features from deviating outside their belonging category clusters. To generate more challenging corrupted point clouds, we adversarially train a noise generator concurrently with the recognition model from the scratch. This differs from previous adversarial training methods that utilized gradient-based attacks as the inner loop. Comprehensive experiments show that the proposed PointCAT outperforms the baseline methods, significantly enhancing the robustness of diverse point cloud recognition models under various corruptions, including isotropic point noises, the LiDAR simulated noises, random point dropping, and adversarial perturbations. Our code is available at: https://github.com/shikiw/PointCAT.

Robust segmenting with noisy labels is an important problem in medical imaging due to the difficulty of acquiring high-quality annotations. Despite the enormous success of recent developments, these developments still require multiple networks to construct their frameworks and focus on limited application scenarios, which leads to inflexibility in practical applications. They also do not explicitly consider the coarse boundary label problem, which results in sub-optimal results. To overcome these challenges, we propose a novel Simultaneous Edge Alignment and Memory-Assisted Learning (SEAMAL) framework for noisy-label robust segmentation. It achieves single-network robust learning, which is applicable for both 2D and 3D segmentation, in both Set-HQ-knowable and Set-HQ-agnostic scenarios. Specifically, to achieve single-model noise robustness, we design a Memory-assisted Selection and Correction module (MSC) that utilizes predictive history consistency from the Prediction Memory Bank to distinguish between reliable and non-reliable labels pixel-wisely, and that updates the reliable ones at the superpixel level. To overcome the coarse boundary label problem, which is common in practice, and to better utilize shape-relevant information at the boundary, we propose an Edge Detection Branch (EDB) that explicitly learns the boundary via an edge detection layer with only slight additional computational cost, and we improve the sharpness and precision of the boundary with a thinning loss. Extensive experiments verify that SEAMAL outperforms previous works significantly.

Pre-training is playing an increasingly important role in learning generic feature representation for Person Re-identification (ReID). We argue that a high-quality ReID representation should have three properties, namely, multi-level awareness, occlusion robustness, and cross-region invariance. To this end, we propose a simple yet effective pre-training framework, namely PersonMAE, which involves two core designs into masked autoencoders to better serve the task of Person Re-ID. 1) PersonMAE generates two regions from the given image withRegionAas the input andRegionBas the prediction target.RegionAis corrupted with block-wise masking to mimic common occlusion in ReID and its remaining visible parts are fed into the encoder. 2) Then PersonMAE aims to predict the wholeRegionBat both pixel level and semantic feature level. It encourages its pre-trained feature representations with the three properties mentioned above. These properties make PersonMAE compatible with downstream Person ReID tasks, leading to state-of-the-art performance on four downstream ReID tasks,i.e.,supervised (holistic and occluded setting), and unsupervised (UDA and USL setting). Notably, on the commonly adopted supervised setting, PersonMAE with ViT-B backbone achieves 79.8% and 69.5% mAP on the MSMT17 and OccDuke datasets, surpassing the previous state-of-the-art by a large margin of +8.0 mAP, and +5.3 mAP, respectively.

Due to the unprecedented power of text-to-image diffusion models, customizing these models to generate new concepts has gained increasing attention. Existing works have achieved some success on real-world concepts, but fail on the concepts of anime characters. We empirically find that such low quality comes from the newly introduced identifier text tokens, which are optimized to identify different characters. In this paper, we proposeAnimeDiffwhich focuses on customized image generation of anime characters. Our AnimeDiff directly binds anime characters with their names and keeps the embeddings of text tokens unchanged. Furthermore, when composing multiple characters in a single image, the model tends to confuse the properties of those characters. To address this issue, our AnimeDiff incorporates aCut-and-Pastedata augmentation strategy that produces multi-character images for training by cutting and pasting multiple characters onto background images. Experiments are conducted to prove the superiority of AnimeDiff over other methods.

Scene graphs offer a structured, hierarchical representation of images, with nodes and edges symbolizing objects and the relationships among them. It can serve as a natural interface for image editing, dramatically improving precision and flexibility. Leveraging this benefit, we introduce a new framework that integrates large language model (LLM) with Text2Image generative model for scene graph-based image editing. This integration enables precise modifications at the object level and creative recomposition of scenes without compromising overall image integrity. Our approach involves two primary stages: 1) Utilizing a LLM-driven scene parser, we construct an image's scene graph, capturing key objects and their interrelationships, as well as parsing fine-grained attributes such as object masks and descriptions. These annotations facilitate concept learning with a fine-tuned diffusion model, representing each object with an optimized token and detailed description prompt. 2) During the image editing phase, a LLM editing controller guides the edits towards specific areas. These edits are then implemented by an attention-modulated diffusion editor, utilizing the fine-tuned model to perform object additions, deletions, replacements, and adjustments. Through extensive experiments, we demonstrate that our framework significantly outperforms existing image editing methods in terms of editing precision and scene aesthetics. Our code is available at https://bestzzhang.github.io/SGEdit.

As a powerful representation of 3D scenes, the neural radiance field (NeRF) enables high-quality novel view synthesis from multi-view images. Stylizing NeRF, however, remains challenging, especially in simulating a text-guided style with both the appearance and the geometry altered simultaneously. In this paper, we present NeRF-Art, a text-guided NeRF stylization approach that manipulates the style of a pre-trained NeRF model with a simple text prompt. Unlike previous approaches that either lack sufficient geometry deformations and texture details or require meshes to guide the stylization, our method can shift a 3D scene to the target style characterized by desired geometry and appearance variations without any mesh guidance. This is achieved by introducing a novel global-local contrastive learning strategy, combined with the directional constraint to simultaneously control both the trajectory and the strength of the target style. Moreover, we adopt a weight regularization method to effectively suppress cloudy artifacts and geometry noises which arise easily when the density field is transformed during geometry stylization. Through extensive experiments on various styles, we demonstrate that our method is effective and robust regarding both single-view stylization quality and cross-view consistency.

Visual question answering (VQA) has experienced tremendous progress in recent years. However, most efforts have only focused on 2D image question-answering tasks. In this article, we extend VQA to its 3D counterpart, 3D question answering (3DQA), which can facilitate a machine's perception of 3D real-world scenarios. Unlike 2D image VQA, 3DQA takes the color point cloud as input and requires both appearance and 3D geometrical comprehension to answer the 3D-related questions. To this end, we propose a novel transformer-based 3DQA framework "3DQA-TR", which consists of two encoders to exploit the appearance and geometry information, respectively. Finally, the multi-modal information about the appearance, geometry, and linguistic question can attend to each other via a 3D-linguistic Bert to predict the target answers. To verify the effectiveness of our proposed 3DQA framework, we further develop the first 3DQA dataset "ScanQA", which builds on the ScanNet dataset and contains over 10 K question-answer pairs for 806 scenes. To the best of our knowledge, ScanQA is the first large-scale dataset with natural-language questions and free-form answers in 3D environments that is fully human-annotated. We also use several visualizations and experiments to investigate the astonishing diversity of the collected questions and the significant differences between this task from 2D VQA and 3D captioning. Extensive experiments on this dataset demonstrate the obvious superiority of our proposed 3DQA framework over state-of-the-art VQA frameworks and the effectiveness of our major designs. Our code and dataset will be made publicly available to facilitate research in this direction. The code and data are available at http://shuquanye.com/3DQA_website/.

This paper explores a better prediction target for BERT pre-training of vision transformers. We observe that current prediction targets disagree with human perception judgment. This contradiction motivates us to learn a perceptual prediction target. We argue that perceptually similar images should stay close to each other in the prediction target space. We surprisingly find one simple yet effective idea: enforcing perceptual similarity during the dVAE training. Moreover, we adopt a self-supervised transformer model for deep feature extraction and show that it works well for calculating perceptual similarity. We demonstrate that such learned visual tokens indeed exhibit better semantic meanings, and help pre-training achieve superior transfer performance in various downstream tasks. For example, we achieve 84.5% Top-1 accuracy on ImageNet-1K with ViT-B backbone, outperforming the competitive method BEiT by +1.3% under the same pre-training epochs. Our approach also gets significant improvement on object detection and segmentation on COCO and semantic segmentation on ADE20K. Equipped with a larger backbone ViT-H, we achieve the state-of-the-art ImageNet accuracy (88.3%) among methods using only ImageNet-1K data.

Diffusion models (DMs) have shown great potential for high-quality image synthesis. However, when it comes to producing images with complex scenes, how to properly describe both image global structures and object details remains a challenging task. In this paper, we present Frido, a Feature Pyramid Diffusion model performing a multi-scale coarse-to-fine denoising process for image synthesis. Our model decomposes an input image into scale-dependent vector quantized features, followed by a coarse-to-fine gating for producing image output. During the above multi-scale representation learning stage, additional input conditions like text, scene graph, or image layout can be further exploited. Thus, Frido can be also applied for conditional or cross-modality image synthesis. We conduct extensive experiments over various unconditioned and conditional image generation tasks, ranging from text-to-image synthesis, layout-to-image, scene-graph-to-image, to label-to-image. More specifically, we achieved state-of-the-art FID scores on five benchmarks, namely layout-to-image on COCO and OpenImages, scene-graph-to-image on COCO and Visual Genome, and label-to-image on COCO.

Human intelligence is multimodal; we integrate visual, linguistic, and acoustic signals to maintain a holistic worldview. Most current pretraining methods, however, are limited to one or two modalities. We present i-Code, a self-supervised pretraining framework where users may flexibly combine the modalities of vision, speech, and language into unified and general-purpose vector representations. In this framework, data from each modality are first given to pretrained single-modality encoders. The encoder outputs are then integrated with a multimodal fusion network, which uses novel merge- and co-attention mechanisms to effectively combine information from the different modalities. The entire system is pretrained end-to-end with new objectives including masked modality unit modeling and cross-modality contrastive learning. Unlike previous research using only video for pretraining, the i-Code framework can dynamically process single, dual, and triple-modality data during training and inference, flexibly projecting different combinations of modalities into a single representation space. Experimental results demonstrate how i-Code can outperform state-of-the-art techniques on five multimodal understanding tasks and single-modality benchmarks, improving by as much as 11% and demonstrating the power of integrative multimodal pretraining.

This paper presents a simple yet effective framework MaskCLIP, which incorporates a newly proposed masked self-distillation into contrastive language-image pretraining. The core idea of masked self-distillation is to distill representation from a full image to the representation predicted from a masked image. Such incorporation enjoys two vital benefits. First, masked self-distillation targets local patch representation learning, which is complementary to vision-language contrastive focusing on text-related representation. Second, masked self-distillation is also consistent with vision-language contrastive from the perspective of training objective as both utilize the visual encoder for feature aligning, and thus is able to learn local semantics getting indirect supervision from the language. We provide specially designed experiments with a comprehensive analysis to validate the two benefits. Symmetrically, we also introduce the local semantic supervision into the text branch, which further improves the pretraining performance. With extensive experiments, we show that MaskCLIP, when applied to various challenging downstream tasks, achieves superior results in linear probing, finetuning, and zeroshot performance with the guidance of the language encoder. Code will be release at https://github.com/LightDXY/MaskCLIP.

We present Diversity-Aware Meta Visual Prompting (DAM-VP), an efficient and effective prompting method for transferring pre-trained models to downstream tasks with frozen backbone. A challenging issue in visual prompting is that image datasets sometimes have a large data diversity whereas a per-dataset generic prompt can hardly handle the complex distribution shift toward the original pretraining data distribution properly. To address this issue, we propose a dataset Diversity-Aware prompting strategy whose initialization is realized by a Meta-prompt. Specifically, we cluster the downstream dataset into small homogeneity subsets in a diversity-adaptive way, with each subset has its own prompt optimized separately. Such a divide-and-conquer design reduces the optimization difficulty greatly and significantly boosts the prompting performance. Furthermore, all the prompts are initialized with a meta-prompt, which is learned across several datasets. It is a bootstrapped paradigm, with the key observation that the prompting knowledge learned from previous datasets could help the prompt to converge faster and perform better on a new dataset. During inference, we dynamically select a proper prompt for each input, based on the feature distance between the input and each subset. Through extensive experiments, our DAM-VP demonstrates superior efficiency and effectiveness, clearly surpassing previous prompting methods in a series of downstream datasets for different pretraining models. Our code is available at: https://github.com/shikiw/DAM-VP.

Combining multiple datasets enables performance boost on many computer vision tasks. But similar trend has not been witnessed in object detection when combining multiple datasets due to two inconsistencies among detection datasets: taxonomy difference and domain gap. In this paper, we address these challenges by a new design (named Detection Hub) that is dataset-aware and category-aligned. It not only mitigates the dataset inconsistency but also provides coherent guidance for the detector to learn across multiple datasets. In particular, the dataset-aware design is achieved by learning a dataset embedding that is used to adapt object queries as well as convolutional kernels in detection heads. The categories across datasets are semantically aligned into a unified space by replacing one-hot category representations with word embedding and leveraging the semantic coherence of language embedding. Detection Hub fulfills the benefits of large data on object detection. Experiments demonstrate that joint training on multiple datasets achieves significant performance gains over training on each dataset alone. Detection Hub further achieves SoTA performance on UODB benchmark with wide variety of datasets.

Benefiting from masked visual modeling, self-supervised video representation learning has achieved remarkable progress. However, existing methods focus on learning representations from scratch through reconstructing low-level features like raw pixel values. In this paper, we propose masked video distillation (MVD), a simple yet effective two-stage masked feature modeling framework for video representation learning: firstly we pretrain an image (or video) model by recovering low-level features of masked patches, then we use the resulting features as targets for masked feature modeling. For the choice of teacher models, we observe that students taught by video teachers perform better on temporally-heavy video tasks, while image teachers transfer stronger spatial representations for spatially-heavy video tasks. Visualization analysis also indicates different teachers produce different learned patterns for students. To leverage the advantage of different teachers, we design a spatial-temporal co-teaching method for MVD. Specifically, we distill student models from both video teachers and image teachers by masked feature modeling. Extensive experimental results demonstrate that video transformers pre-trained with spatial-temporal co-teaching outperform models distilled with a single teacher on a multitude of video datasets. Our MVD with vanilla ViT achieves state-of-the-art performance compared with previous methods on several challenging video downstream tasks. For example, with the ViT-Large model, our MVD achieves 86.4% and 76.7% Top-1 accuracy on Kinetics-400 and Something-Something-v2, outperforming VideoMAE by 1.2% and 2.4% respectively. When a larger ViT-Huge model is adopted, MVD achieves the state-of-the-art performance with 77.3% Top-1 accuracy on Something-Something-v2. Code will be available at https://github.com/ruiwang2021/mvd.

Exploring dense matching between the current frame and past frames for long-range context modeling, memory-based methods have demonstrated impressive results in video object segmentation (VOS) recently. Nevertheless, due to the lack of instance understanding ability, the above approaches are oftentimes brittle to large appearance variations or viewpoint changes resulted from the movement of objects and cameras. In this paper, we argue that instance understanding matters in VOS, and integrating it with memory-based matching can enjoy the synergy, which is intuitively sensible from the definition of VOS task, i.e., identifying and segmenting object instances within the video. Towards this goal, we present a two-branch network for VOS, where the query-based instance segmentation (IS) branch delves into the instance details of the current frame and the VOS branch performs spatial-temporal matching with the memory bank. We employ the well-learned object queries from IS branch to inject instance-specific information into the query key, with which the instance-augmented matching is further performed. In addition, we introduce a multi-path fusion block to effectively combine the memory readout with multi-scale features from the instance segmentation decoder, which incorporates high-resolution instance-aware features to produce final segmentation results. Our method achieves state-of-the-art performance on DAVIS 2016/2017 val (92.6% and 87.1%), DAVIS 2017 test-dev (82.8%), and YouTube-VOS 2018/2019 val (86.3% and 86.3%), outperforming alternative methods by clear margins.

This paper focuses on analyzing and improving the commonsense ability of recent popular vision-language (VL) models. Despite the great success, we observe that existing VL-models still lack commonsense knowledge/reasoning ability (e.g., “Lemons are sour”), which is a vital component towards artificial general intelligence. Through our analysis, we find one important reason is that existing large-scale VL datasets do not contain much commonsense knowledge, which motivates us to improve the commonsense of VL-models from the data perspective. Rather than collecting a new VL training dataset, we propose a more scalable strategy, i.e., “Data Augmentation with kNowledge graph linearization for CommonsensE capability” (DANCE). It can be viewed as one type of data augmentation technique, which can inject commonsense knowledge into existing VL datasets on the fly during training. More specifically, we leverage the commonsense knowledge graph (e.g., ConceptNet) and create variants of text description in VL datasets via bidirectional sub-graph sequentialization. For better commonsense evaluation, we further propose the first retrieval-based commonsense diagnostic benchmark. By conducting extensive experiments on some representative VL-models, we demonstrate that our DANCE technique is able to significantly improve the commonsense ability while maintaining the performance on vanilla retrieval tasks. The code and data are available at https://github.com/pleaseconnectwifi/DANCE.

Video understanding tasks have traditionally been modeled by two separate architectures, specially tailored for two distinct tasks. Sequence-based video tasks, such as action recognition, use a video backbone to directly extract spatiotemporal features, while frame-based video tasks, such as multiple object tracking (MOT), rely on single fixed-image backbone to extract spatial features. In contrast, we propose to unify video understanding tasks into one novel streaming video architecture, referred to as Streaming Vision Transformer (S-ViT). S-ViT first produces frame-level features with a memory-enabled temporally-aware spatial encoder to serve the frame-based video tasks. Then the frame features are input into a task-related temporal decoder to obtain spatiotemporal features for sequence-based tasks. The efficiency and efficacy of S-ViT is demonstrated by the state-of-the-art accuracy in the sequence-based action recognition task and the competitive advantage over conventional architecture in the frame-based MOT task. We believe that the concept of streaming video model and the implementation of S-ViT are solid steps towards a unified deep learning architecture for video understanding. Code will be available at https://github.com/yuzhms/Streaming-Video-Model.

In this paper, we investigate the adversarial robustness of vision transformers that are equipped with BERT pretraining (e.g., BEiT, MAE). A surprising observation is that MAE has significantly worse adversarial robustness than other BERT pretraining methods. This observation drives us to rethink the basic differences between these BERT pretraining methods and how these differences affect the robustness against adversarial perturbations. Our empirical analysis reveals that the adversarial robustness of BERT pretraining is highly related to the reconstruction target, i.e., predicting the raw pixels of masked image patches will degrade more adversarial robustness of the model than predicting the semantic context, since it guides the model to concentrate more on medium-/high-frequency components of images. Based on our analysis, we provide a simple yet effective way to boost the adversarial robustness of MAE. The basic idea is using the dataset-extracted domain knowledge to occupy the medium-/high-frequency of images, thus narrowing the optimization space of adversarial perturbations. Specifically, we group the distribution of pretraining data and optimize a set of cluster-specific visual prompts on frequency domain. These prompts are incorporated with input images through prototype-based prompt selection during test period. Extensive evaluation shows that our method clearly boost MAE’s adversarial robustness while maintaining its clean performance on ImageNet-1k classification. Our code is available at: https://github.com/shikiw/RobustMAE.

Neural implicit fields are powerful for representing 3D scenes and generating high-quality novel views, but it remains challenging to use such implicit representations for creating a 3D human avatar with a specific identity and artistic style that can be easily animated. Our proposed method, AvatarCraft, addresses this challenge by using diffusion models to guide the learning of geometry and texture for a neural avatar based on a single text prompt. We carefully design the optimization framework of neural implicit fields, including a coarse-to-fine multi-bounding box training strategy, shape regularization, and diffusion-based constraints, to produce high-quality geometry and texture. Additionally, we make the human avatar animatable by deforming the neural implicit field with an explicit warping field that maps the target human mesh to a template human mesh, both represented using parametric human models. This simplifies animation and reshaping of the generated avatar by controlling pose and shape parameters. Extensive experiments on various text descriptions show that AvatarCraft is effective and robust in creating human avatars and rendering novel views, poses, and shapes. Our project page is: https://avatar-craft.github.io/.

Hair editing has made tremendous progress in recent years. Early hair editing methods use well-drawn sketches or masks to specify the editing conditions. Even though they can enable very fine-grained local control, such interaction modes are inefficient for the editing conditions that can be easily specified by language descriptions or reference images. Thanks to the recent breakthrough of cross-modal models (e.g., CLIP), HairCLIP is the first work that enables hair editing based on text descriptions or reference images. However, such text-driven and reference-driven interaction modes make HairCLIP unable to support fine-grained controls specified by sketch or mask. In this paper, we propose HairCLIPv2, aiming to support all the aforementioned interactions with one unified framework. Simultaneously, it improves upon HairCLIP with better irrelevant attributes (e.g., identity, background) preservation and unseen text descriptions support. The key idea is to convert all the hair editing tasks into hair transfer tasks, with editing conditions converted into different proxies accordingly. The editing effects are added upon the input image by blending the corresponding proxy features within the hairstyle or hair color feature spaces. Besides the unprecedented user interaction mode support, quantitative and qualitative experiments demonstrate the superiority of HairCLIPv2 in terms of editing effects, irrelevant attribute preservation and visual naturalness. Our code is available at https://github.com/wty-ustc/HairCLIPv2.

Copy-Paste is a simple and effective data augmentation strategy for instance segmentation. By randomly pasting object instances onto new background images, it creates new training data for free and significantly boosts the segmentation performance, especially for rare object categories. Although diverse, high-quality object instances used in Copy-Paste result in more performance gain, previous works utilize object instances either from human-annotated instance segmentation datasets or rendered from 3D object models, and both approaches are too expensive to scale up to obtain good diversity. In this paper, we revisit Copy-Paste at scale with the power of newly emerged zero-shot recognition models (e.g., CLIP) and text2image models (e.g., StableDiffusion). We demonstrate for the first time that using a text2image model to generate images or zero-shot recognition model to filter noisily crawled images for different object categories is a feasible way to make Copy-Paste truly scalable. To make such success happen, we design a data acquisition and processing framework, dubbed ``X-Paste", upon which a systematic study is conducted. On the LVIS dataset, X-Paste provides impressive improvements over the strong baseline CenterNet2 with Swin-L as the backbone. Specifically, it archives +2.6 box AP and +2.1 mask AP gains on all classes and even more significant gains with +6.8 box AP +6.5 mask AP on long-tail classes.

Long-tailed object detection (LTOD) aims to handle the extreme data imbalance in real-world datasets, where many tail classes have scarce instances. One popular strategy is to explore extra data with image-level labels, yet it produces limited results due to (1) semantic ambiguity---an image-level label only captures a salient part of the image, ignoring the remaining rich semantics within the image; and (2) location sensitivity---the label highly depends on the locations and crops of the original image, which may change after data transformations like random cropping. To remedy this, we propose RichSem, a simple but effective method, which is robust to learn rich semantics from coarse locations without the need of accurate bounding boxes. RichSem leverages rich semantics from images, which are then served as additional ``soft supervision'' for training detectors. Specifically, we add a semantic branch to our detector to learn these soft semantics and enhance feature representations for long-tailed object detection. The semantic branch is only used for training and is removed during inference. RichSem achieves consistent improvements on both overall and rare-category of LVIS under different backbones and detectors. Our method achieves state-of-the-art performance without requiring complex training and testing procedures. Moreover, we show the effectiveness of our method on other long-tailed datasets with additional experiments.

Text-to-Image diffusion models have made tremendous progress over the past two years, enabling the generation of highly realistic images based on open-domain text descriptions. However, despite their success, text descriptions often struggle to adequately convey detailed controls, even when composed of long and complex texts. Moreover, recent studies have also shown that these models face challenges in understanding such complex texts and generating the corresponding images. Therefore, there is a growing need to enable more control modes beyond text description. In this paper, we introduce Uni-ControlNet, a unified framework that allows for the simultaneous utilization of different local controls (e.g., edge maps, depth map, segmentation masks) and global controls (e.g., CLIP image embeddings) in a flexible and composable manner within one single model. Unlike existing methods, Uni-ControlNet only requires the fine-tuning of two additional adapters upon frozen pre-trained text-to-image diffusion models, eliminating the huge cost of training from scratch. Moreover, thanks to some dedicated adapter designs, Uni-ControlNet only necessitates a constant number (i.e., 2) of adapters, regardless of the number of local or global controls used. This not only reduces the fine-tuning costs and model size, making it more suitable for real-world deployment, but also facilitate composability of different conditions. Through both quantitative and qualitative comparisons, Uni-ControlNet demonstrates its superiority over existing methods in terms of controllability, generation quality and composability. Code is available at https://github.com/ShihaoZhaoZSH/Uni-ControlNet.

Unsupervised domain adaption has been widely adopted in tasks with scarce annotated data. Unfortunately, mapping the target-domain distribution to the source-domain unconditionally may distort the essential structural information of the target-domain data, leading to inferior performance. To address this issue, we first propose to introduce active sample selection to assist domain adaptation regarding the semantic segmentation task. By innovatively adopting multiple anchors instead of a single centroid, both source and target domains can be better characterized as multimodal distributions, in which way more complementary and informative samples are selected from the target domain. With only a little workload to manually annotate these active samples, the distortion of the target-domain distribution can be effectively alleviated, achieving a large performance gain. In addition, a powerful semi-supervised domain adaptation strategy is proposed to alleviate the long-tail distribution problem and further improve the segmentation performance. Extensive experiments are conducted on public datasets, and the results demonstrate that the proposed approach outperforms state-of-the-art methods by large margins and achieves similar performance to the fully-supervised upperbound, i.e., 71.4% mIoU on GTA5 and 71.8% mIoU on SYNTHIA. The effectiveness of each component is also verified by thorough ablation studies.

Semantic image synthesis, translating semantic layouts to photo-realistic images, is a one-to-many mapping problem. Though impressive progress has been recently made, diverse semantic synthesis that can efficiently produce semantic-level or even instance-level multimodal results, still remains a challenge. In this article, we propose a novel diverse semantic image synthesis framework from the perspective of semantic class distributions, which naturally supports diverse generation at both semantics and instance level. We achieve this by modeling class-level conditional modulation parameters as continuous probability distributions instead of discrete values, and sampling per-instance modulation parameters through instance-adaptive stochastic sampling that is consistent across the network. Moreover, we propose prior noise remapping, through linear perturbation parameters encoded from paired references, to facilitate supervised training and exemplar-based instance style control at test time. To further extend the user interaction function of the proposed method, we also introduce sketches into the network. In addition, specially designed generator modules, Progressive Growing Module and Multi-Scale Refinement Module, can be used as a general module to improve the performance of complex scene generation. Extensive experiments on multiple datasets show that our method can achieve superior diversity and comparable quality compared to state-of-the-art methods. Codes are available at https://github.com/tzt101/INADE.git.

We propose to restore old photos that suffer from severe degradation through a deep learning approach. Unlike conventional restoration tasks that can be solved through supervised learning, the degradation in real photos is complex and the domain gap between synthetic images and real old photos makes the network fail to generalize. Therefore, we propose a novel triplet domain translation network by leveraging real photos along with massive synthetic image pairs. Specifically, we train two variational autoencoders (VAEs) to respectively transform old photos and clean photos into two latent spaces. And the translation between these two latent spaces is learned with synthetic paired data. This translation generalizes well to real photos because the domain gap is closed in the compact latent space. Besides, to address multiple degradations mixed in one old photo, we design a global branch with a partial nonlocal block targeting the structured defects, such as scratches and dust spots, and a local branch targeting the unstructured defects, such as noises and blurriness. We also extend the global branch with a more memory-efficient scheme, named multi-scale patch-based attention to processing high-resolution photos. Two branches are fused in the latent space, leading to improved capability to restore old photos from multiple defects. Furthermore, we apply another face refinement network to recover fine details of faces in the old photos, thus ultimately generating photos with enhanced perceptual quality. With comprehensive experiments, the proposed pipeline demonstrates superior performance over state-of-the-art methods as well as existing commercial tools in terms of visual quality for old photos restoration. Both code and models could be found at https://github.com/microsoft/Bringing-Old-Photos-Back-to-Life.

Point cloud segmentation is a fundamental task in 3D. Despite recent progress on point cloud segmentation with the power of deep networks, current learning methods based on the clean label assumptions may fail with noisy labels. Yet, class labels are often mislabeled at both instance-level and boundary-level in real-world datasets. In this work, we take the lead in solving the instance-level label noise by proposing a Point Noise-Adaptive Learning (PNAL) framework. Compared to noise-robust methods on image tasks, our framework is noise-rate blind, to cope with the spatially variant noise rate specific to point clouds. Specifically, we propose a point-wise confidence selection to obtain reliable labels from the historical predictions of each point. A cluster-wise label correction is proposed with a voting strategy to generate the best possible label by considering the neighbor correlations. To handle boundary-level label noise, we also propose a variant "PNAL-boundary " with a progressive boundary label cleaning strategy. Extensive experiments demonstrate its effectiveness on both synthetic and real-world noisy datasets. Even with 60% symmetric noise and high-level boundary noise, our framework significantly outperforms its baselines, and is comparable to the upper bound trained on completely clean data. Moreover, we cleaned the popular real-world dataset ScanNetV2 for rigorous experiment. Our code and data is available at https://github.com/pleaseconnectwifi/PNAL.

In recent years, many model intellectual property (IP) proof methods for IP protection have been proposed, such as model watermarking and model fingerprinting. However, with the increasing number of models transmitted and deployed on the Internet, quickly finding the suspect model among thousands of models on model-sharing platforms such as GitHub is in great demand, which concurrently triggers the new security problem of model copy detection for IP protection. As an important part of the model IP protection system, the model copy detection task has not received enough attention. Due to the high computational complexity, both model watermarking and model fingerprinting lack the capability to efficiently find suspected infringing models among tens of millions of models. In this article, inspired by the hash-based image retrieval methods, we introduce a novel model copy detection mechanism: perceptual hashing for convolutional neural networks (CNNs). The proposed perceptual hashing algorithm can convert the weights of CNNs to fixed-length binary hash codes so that the lightly modified version has the similar hash code as the original model. By comparing the similarity of a pair of hash codes between a query model and a test model in the model library, similar versions of a query model can be retrieved efficiently. To the best of our knowledge, this is the first perceptual hashing algorithm for deep neural network models. Specifically, we first select the important model weights based on the model compression theory, then calculate the normal test statistics (NTS) on the segments of important weights, and finally encode the NTS features into hash codes. The experiment performed on a model library containing 3,565 models indicates that our perceptual hashing scheme has a superior copy detection performance.

Face image manipulation via three-dimensional guidance has been widely applied in various interactive scenarios due to its semantically-meaningful understanding and user-friendly controllability. However, existing 3D-morphable-model-based manipulation methods are not directly applicable to out-of-domain faces, such as non-photorealistic paintings, cartoon portraits, or even animals, mainly due to the formidable difficulties in building the model for each specific face domain. To overcome this challenge, we propose, as far as we know, the first method to manipulate faces in arbitrary domains using human 3DMM. This is achieved through two major steps: 1) disentangled mapping from 3DMM parameters to the latent space embedding of a pre-trained StyleGAN2 [1] that guarantees disentangled and precise controls for each semantic attribute; and 2) cross-domain adaptation that bridges domain discrepancies and makes human 3DMM applicable to out-of-domain faces by enforcing a consistent latent space embedding. Experiments and comparisons demonstrate the superiority of our high-quality semantic manipulation method on a variety of face domains with all major 3D facial attributes controllable - pose, expression, shape, albedo, and illumination. Moreover, we develop an intuitive editing interface to support user-friendly control and instant feedback. Our project page is https://cassiepython.github.io/cddfm3d/index.html.

We present Mobile-Former, a parallel design of MobileNet and transformer with a two-way bridge in between. This structure leverages the advantages of MobileNet at local processing and transformer at global interaction. And the bridge enables bidirectional fusion of local and global features. Different from recent works on vision transformer, the transformer in Mobile-Former contains very few tokens (e.g. 6 or fewer tokens) that are randomly initialized to learn global priors, resulting in low computational cost. Combining with the proposed light-weight cross attention to model the bridge, Mobile-Former is not only computationally efficient, but also has more representation power. It outperforms MobileNetV3 at low FLOP regime from 25M to 500M FLOPs on ImageNet classification. For instance, Mobile-Former achieves 77.9% top-1 accuracy at 294M FLOPs, gaining 1.3% over MobileNetV3 but saving 17% of computations. When transferring to object detection, Mobile-Former outperforms MobileNetV3 by 8.6 AP in RetinaNet framework. Furthermore, we build an efficient end-to-end detector by replacing backbone, encoder and decoder in DETR with Mobile-Former, which outperforms DETR by 1.3 AP but saves 52% of computational cost and 36% of parameters. Code will be released at https://github.com/aaboys/mobileformer.

In this work we propose Identity Consistency Transformer, a novel face forgery detection method that focuses on high-level semantics, specifically identity information, and detecting a suspect face by finding identity inconsistency in inner and outer face regions. The Identity Consistency Transformer incorporates a consistency loss for identity consistency determination. We show that Identity Consistency Transformer exhibits superior generalization ability not only across different datasets but also across various types of image degradation forms found in real-world applications including deepfake videos. The Identity Consistency Transformer can be easily enhanced with additional identity information when such information is available, and for this reason it is especially well-suited for detecting face forgeries involving celebrities.11Code will be released at https://github.com/LightDXY/ICT_DeepFake

We present CSWin Transformer, an efficient and effective Transformer-based backbone for general-purpose vision tasks. A challenging issue in Transformer design is that global self-attention is very expensive to compute whereas local self-attention often limits the field of interactions of each token. To address this issue, we develop the Cross-Shaped Window self-attention mechanism for computing self-attention in the horizontal and vertical stripes in parallel that form a cross-shaped window, with each stripe obtained by splitting the input feature into stripes of equal width. We provide a mathematical analysis of the effect of the stripe width and vary the stripe width for different layers of the Transformer network which achieves strong modeling capability while limiting the computation cost. We also introduce Locally-enhanced Positional Encoding (LePE), which handles the local positional information better than existing encoding schemes. LePE naturally supports arbitrary input resolutions, and is thus especially effective and friendly for downstream tasks. Incorporated with these designs and a hierarchical structure, CSWin Transformer demonstrates competitive performance on common vision tasks. Specifically, it achieves 85.4% Top-1 accuracy on ImageNet-1K without any extra training data or label, 53.9 box AP and 46.4 mask AP on the COCO detection task, and 52.2 mIOU on the ADE20K semantic segmentation task, surpassing previous state-of-the-art Swin Transformer backbone by +1.2, +2.0, +1.4, and +2.0 respectively under the similar FLOPs setting. By further pretraining on the larger dataset ImageNet-21K, we achieve 87.5% Top-1 accuracy on ImageNet-1K and high segmentation performance on ADE20K with 55.7 mIoU.11Code and pretrain model is available at https://github.com/microsoft/CSWin-Transformer

This paper aims to address the problem of pretraining for person re-identification (Re-ID) with noisy labels. To setup the pretraining task, we apply a simple online multi-object tracking system on raw videos of an existing un-labeled Re-ID dataset “LUPerson” and build the Noisy Labeled variant called “LUPerson-NL”. Since theses ID labels automatically derived from tracklets inevitably con-tain noises, we develop a large-scale Pre-training frame-work utilizing Noisy Labels (PNL), which consists of three learning modules: supervised Re-ID learning, prototype-based contrastive learning, and label-guided contrastive learning. In principle, joint learning of these three mod-ules not only clusters similar examples to one prototype, but also rectifies noisy labels based on the prototype as-signment. We demonstrate that learning directly from raw videos is a promising alternative for pre-training, which utilizes spatial and temporal correlations as weak super-vision. This simple pre-training task provides a scalable way to learn SOTA Re-ID representations from scratch on “LUPerson-NL” without bells and whistles. For example, by applying on the same supervised Re-ID method MGN, our pre-trained model improves the mAP over the unsu-pervised pre-training counterpart by 5.7%, 2.2%, 2.3% on CUHK03, DukeMTMC, and MSMT17 respectively. Under the small-scale or few-shot setting, the performance gain is even more significant, suggesting a better transferability of the learned representation. Code is available at https://github.com/DengpanFu/LUPerson-NL.

We present the vector quantized diffusion (VQ-Diffusion) model for text-to-image generation. This method is based on a vector quantized variational autoencoder (VQ-VAE) whose latent space is modeled by a conditional variant of the recently developed Denoising Diffusion Probabilistic Model (DDPM). We find that this latent-space method is well-suited for text-to-image generation tasks because it not only eliminates the unidirectional bias with existing methods but also allows us to incorporate a mask-and-replace diffusion strategy to avoid the accumulation of errors, which is a serious problem with existing methods. Our experiments show that the VQ-Diffusion produces significantly better text-to-image generation results when compared with conventional autoregressive (AR) models with similar numbers of parameters. Compared with previous GAN-based text-to-image methods, our VQ-Diffusion can handle more complex scenes and improve the synthesized image quality by a large margin. Finally, we show that the image generation computation in our method can be made highly efficient by reparameterization. With traditional AR methods, the text-to-image generation time increases linearly with the output image resolution and hence is quite time consuming even for normal size images. The VQ-Diffusion allows us to achieve a better trade-off between quality and speed. Our experiments indicate that the VQ-Diffusion model with the reparameterization is fifteen times faster than traditional AR methods while achieving a better image quality. The code and models are available at https://github.com/cientgu/VQ-Diffusion.

Adversary and invisibility are two fundamental but conflict characters of adversarial perturbations. Previous adversarial attacks on 3D point cloud recognition have often been criticized for their noticeable point outliers, since they just involve an “implicit constrain” like global distance loss in the time-consuming optimization to limit the generated noise. While point cloud is a highly structured data format, it is hard to constrain its perturbation with a simple loss or metric properly. In this paper, we propose a novel Point-Cloud Sensitivity Map to boost both the efficiency and imperceptibility of point perturbations. This map reveals the vulnerability of point cloud recognition models when encountering shape-invariant adversarial noises. These noises are designed along the shape surface with an “explicit constrain” instead of extra distance loss. Specifically, we first apply a reversible coordinate transformation on each point of the point cloud input, to reduce one degree of point freedom and limit its movement on the tangent plane. Then we calculate the best attacking direction with the gradients of the transformed point cloud obtained on the white-box model. Finally we assign each point with a non-negative score to construct the sensitivity map, which benefits both white-box adversarial invisibility and black-box query-efficiency extended in our work. Extensive evaluations prove that our method can achieve the superior performance on various point cloud recognition models, with its satisfying adversarial imperceptibility and strong resistance to different point cloud defense settings. Our code is available at: https://github.com/shikiw/SI-Adv.

Transformers have achieved great success in pluralistic image inpainting recently. However, we find existing transformer based solutions regard each pixel as a token, thus suffer from information loss issue from two aspects: 1) They downsample the input image into much lower resolutions for efficiency consideration, incurring information loss and extra misalignment for the boundaries of masked regions. 2) They quantize 2563RGB pixels to a small number (such as 512) of quantized pixels. The indices of quantized pixels are used as tokens for the inputs and prediction targets of transformer. Although an extra CNN network is used to upsample and refine the low-resolution results, it is difficult to retrieve the lost information back. To keep input information as much as possible, we propose a new transformer based framework “PUT”. Specifically, to avoid input downsampling while maintaining the computation efficiency, we design a patch-based auto-encoder P-VQVAE, where the encoder converts the masked image into non-overlapped patch tokens and the decoder recovers the masked regions from the inpainted tokens while keeping the unmasked regions unchanged. To eliminate the information loss caused by quantization, an Un-Quantized Transformer (UQ-Transformer) is applied, which directly takes the features from P-VQVAE encoder as input without quantization and regards the quantized tokens only as prediction targets. Extensive experiments show that PUT greatly outperforms state-of-the-art methods on image fidelity, especially for large masked regions and complex large-scale datasets.

We present a learning-based framework, recurrent transformer network (RTN), to restore heavily degraded old films. Instead of performing frame-wise restoration, our method is based on the hidden knowledge learned from adjacent frames that contain abundant information about the occlusion, which is beneficial to restore challenging artifacts of each frame while ensuring temporal coherency. Moreover, contrasting the representation of the current frame and the hidden knowledge makes it possible to infer the scratch position in an unsupervised manner, and such defect localization generalizes well to real-world degradations. To better resolve mixed degradation and compensate for the flow estimation error during frame alignment, we propose to leverage more expressive transformer blocks for spatial restoration. Experiments on both synthetic dataset and real-world old films demonstrate the significant superiority of the proposed RTN over existing solutions. In addition, the same framework can effectively propagate the color from keyframes to the whole video, ultimately yielding compelling restored films. The implementation and model will be released at https://github.com/raywzy/Bringing-Old-Films-Back-to-Life.

We present CLIP-NeRF, a multi-modal 3D object manipulation method for neural radiance fields (NeRF). By leveraging the joint language-image embedding space of the recent Contrastive Language-Image Pre-Training (CLIP) model, we propose a unified framework that allows manip-ulating NeRF in a user-friendly way, using either a short text prompt or an exemplar image. Specifically, to combine the novel view synthesis capability of NeRF and the controllable manipulation ability of latent representations from generative models, we introduce a disentangled conditional NeRF architecture that allows individual control over both shape and appearance. This is achieved by performing the shape conditioning via applying a learned deformation field to the positional encoding and deferring color conditioning to the volumetric rendering stage. To bridge this disentangled latent representation to the CLIP embedding, we design two code mappers that take a CLIP embedding as input and update the latent codes to reflect the targeted editing. The mappers are trained with a CLIP-based matching loss to ensure the manipulation accuracy. Furthermore, we propose an inverse optimization method that accurately projects an input image to the latent codes for manipulation to enable editing on real images. We evaluate our approach by extensive experiments on a variety of text prompts and exemplar images and also provide an intuitive interface for interactive editing.

This paper studies the BERT pretraining of video transformers. It is a straightforward but worth-studying extension given the recent success from BERT pretraining of image transformers. We introduce BEVT which decouples video representation learning into spatial representation learning and temporal dynamics learning. In particular, BEVT first performs masked image modeling on image data, and then conducts masked image modeling jointly with masked video modeling on video data. This design is motivated by two observations: 1) transformers learned on image datasets provide decent spatial priors that can ease the learning of video transformers, which are often times computationally-intensive if trained from scratch; 2) discriminative clues, i.e., spatial and temporal information, needed to make correct predictions vary among different videos due to large intra-class and inter-class variations. We conduct extensive experiments on three challenging video benchmarks where BEVT achieves very promising results. On Kinetics 400, for which recognition mostly relies on discriminative spatial representations, BEVT achieves comparable results to strong supervised baselines. On Something-Something-V2 and Diving 48, which contain videos relying on temporal dynamics, BEVT outperforms by clear margins all alternative baselines and achieves state-of-the-art performance with a 71.4% and 87.2% Top-1 accuracy respectively. Code is available at https://github.com/xyzforever/BEVT.

Hair editing is an interesting and challenging problem in computer vision and graphics. Many existing methods require well-drawn sketches or masks as conditional inputs for editing, however these interactions are neither straight-forward nor efficient. In order to free users from the tedious interaction process, this paper proposes a new hair editing interaction mode, which enables manipulating hair attributes individually or jointly based on the texts or reference images provided by users. For this purpose, we encode the image and text conditions in a shared embedding space and propose a unified hair editing framework by leveraging the powerful image text representation capability of the Contrastive Language-Image Pre-Training (CLIP) model. With the carefully designed network structures and loss functions, our framework can perform high-quality hair editing in a disentangled manner. Extensive experiments demonstrate the superiority of our approach in terms of manipulation accuracy, visual realism of editing results, and irrelevant attribute preservation.

How to learn a universal facial representation that boosts all face analysis tasks? This paper takes one step toward this goal. In this paper, we study the transfer performance of pre-trained models on face analysis tasks and introduce a framework, called FaRL, for general facial representation learning. On one hand, the framework involves a contrastive loss to learn high-level semantic meaning from image-text pairs. On the other hand, we propose exploring low-level information simultaneously to further enhance the face representation by adding a masked image modeling. We perform pre-training on LAION-FACE, a dataset containing a large amount of face image-text pairs, and evaluate the representation capability on multiple downstream tasks. We show that FaRL achieves better transfer performance compared with previous pre-trained models. We also verify its superiority in the low-data regime. More importantly, our model surpasses the state-of-the-art methods on face analysis tasks including face parsing and face alignment.

This paper revisits visual representation in knowledge-based visual question answering (VQA) and demonstrates that using regional information in a better way can significantly improve the performance. While visual representation is extensively studied in traditional VQA, it is under-explored in knowledge-based VQA even though these two tasks share the common spirit, i.e., rely on visual input to answer the question. Specifically, we observe in most state-of-the-art knowledge-based VQA methods: 1) visual features are extracted either from the whole image or in a sliding window manner for retrieving knowledge, and the important relationship within/among object regions is neglected; 2) visual features are not well utilized in the final answering model, which is counter-intuitive to some extent. Based on these observations, we propose a new knowledge-based VQA method REVIVE, which tries to utilize the explicit information of object regions not only in the knowledge retrieval stage but also in the answering model. The key motivation is that object regions and inherent relationship are important for knowledge-based VQA. We perform extensive experiments on the standard OK-VQA dataset and achieve new state-of the-art performance, i.e., 58.0 accuracy, surpassing previous state-of-the-art method by a large margin (+3.6%). We also conduct detailed analysis and show the necessity of regional information in different framework components for knowledge-based VQA. Code is publicly available at https://github.com/yzleroy/REVIVE.

This paper presents OmniVL, a new foundation model to support both image-language and video-language tasks using one universal architecture. It adopts a unified transformer-based visual encoder for both image and video inputs, and thus can perform joint image-language and video-language pretraining. We demonstrate, for the first time, such a paradigm benefits both image and video tasks, as opposed to the conventional one-directional transfer (e.g., use image-language to help video-language). To this end, we propose a \emph{decoupled} joint pretraining of image-language and video-language to effectively decompose the vision-language modeling into spatial and temporal dimensions and obtain performance boost on both image and video tasks. Moreover, we introduce a novel unified vision-language contrastive (UniVLC) loss to leverage image-text, video-text, image-label (e.g., image classification), video-label (e.g., video action recognition) data together, so that both supervised and noisily supervised pretraining data are utilized as much as possible. Without incurring extra task-specific adaptors, OmniVL can simultaneously support visual only tasks (e.g., image classification, video action recognition), cross-modal alignment tasks (e.g., image/video-text retrieval), and multi-modal understanding and generation tasks (e.g., image/video question answering, captioning). We evaluate OmniVL on a wide range of downstream tasks and achieve state-of-the-art or competitive results with similar model size and data scale.

Spatially-adaptive normalization (SPADE) is remarkably successful recently in conditional semantic image synthesis in T. Park et al. 2019 which modulates the normalized activation with spatially-varying transformations learned from semantic layouts, to prevent the semantic information from being washed away. Despite its impressive performance, a more thorough understanding of the advantages inside the box is still highly demanded to help reduce the significant computation and parameter overhead introduced by this novel structure. In this paper, from a return-on-investment point of view, we conduct an in-depth analysis of the effectiveness of this spatially-adaptive normalization and observe that its modulation parameters benefit more from semantic-awareness rather than spatial-adaptiveness, especially for high-resolution input masks. Inspired by this observation, we propose class-adaptive normalization (CLADE), a lightweight but equally-effective variant that is only adaptive to semantic class. In order to further improve spatial-adaptiveness, we introduce intra-class positional map encoding calculated from semantic layouts to modulate the normalization parameters of CLADE and propose a truly spatially-adaptive variant of CLADE, namely CLADE-ICPE. Through extensive experiments on multiple challenging datasets, we demonstrate that the proposed CLADE can be generalized to different SPADE-based methods while achieving comparable generation quality compared to SPADE, but it is much more efficient with fewer extra parameters and lower computational cost. The code and pretrained models are available at https://github.com/tzt101/CLADE.git.

Despite the tremendous success, deep neural networks are exposed to serious IP infringement risks. Given a target deep model, if the attacker knows its full information, it can be easily stolen by fine-tuning. Even if only its output is accessible, a surrogate model can be trained through student-teacher learning by generating many input-output training pairs. Therefore, deep model IP protection is important and necessary. However, it is still seriously under-researched. In this work, we propose a new model watermarking framework for protecting deep networks trained for low-level computer vision or image processing tasks. Specifically, a special task-agnostic barrier is added after the target model, which embeds a unified and invisible watermark into its outputs. When the attacker trains one surrogate model by using the input-output pairs of the barrier target model, the hidden watermark will be learned and extracted afterwards. To enable watermarks from binary bits to high-resolution images, a deep invisible watermarking mechanism is designed. By jointly training the target model and watermark embedding, the extra barrier can even be absorbed into the target model. Through extensive experiments, we demonstrate the robustness of the proposed framework, which can resist attacks with different network structures and objective functions.

Steganography is the art and science of hiding secret messages in public communication so that the presence of secret messages cannot be detected. There are two distribution-preserving steganographic frameworks, one is sampler-based and the other is compression-based. The former requires a perfect sampler which yields data following the same distribution, and the latter needs the explicit distribution of generative objects. However, these two conditions are too strict even unrealistic in the traditional data environment, e.g., the distribution of natural images is hard to seize. Fortunately, generative models bring new vitality to distribution-preserving steganography, which can serve as the perfect sampler or provide the explicit distribution of generative media. Taking text-to-speech generation task as an example, we propose distribution-preserving steganography based on WaveGlow and WaveRNN, which corresponds to the former two categories. Steganalysis experiments and theoretical analysis are conducted to demonstrate that the proposed methods can preserve the distribution.

Automatic translation of aerial images into digital maps is an important and challenging task which is widely used in practical applications. Most of the existing works view it either as a creative image-to-image translation problem or a discriminative semantic segmentation problem. However, we notice that human annotators need to extract and understand the information in aerial images first and then translate them to online maps in a creative way, which helps them draw accurate and visually appealing online maps. In this article, we propose an end-to-end online map generation method that combines a discriminative module with a creative module based on this observation to mimic human behavior. Specifically, we first utilize a semantic segmentation module to obtain a rough aerial map, in which each region is labeled with its category, and then further improve its quality with a creative module. To train a robust network that generalizes well to unfamiliar regions, we also collect a large aerial image dataset for online map generation (AIDOMG). AIDOMG consists of 40 087 pairs of aerial images and corresponding online maps collected from nine regions of six continents. We conduct extensive experiments to verify the superiority of the new design that combines discrimination and creativity and experimental results show that the performance of the proposed method significantly outperforms baseline methods.

This paper studies the problem of StyleGAN inversion, which plays an essential role in enabling the pretrained StyleGAN to be used for real image editing tasks. The goal of StyleGAN inversion is to find the exact latent code of the given image in the latent space of StyleGAN. This problem has a high demand for quality and efficiency. Existing optimization-based methods can produce high-quality results, but the optimization often takes a long time. On the contrary, forward-based methods are usually faster but the quality of their results is inferior. In this paper, we present a new feed-forward network "E2Style" for StyleGAN inversion, with significant improvement in terms of efficiency and effectiveness. In our inversion network, we introduce: 1) a shallower backbone with multiple efficient heads across scales; 2) multi-layer identity loss and multi-layer face parsing loss to the loss function; and 3) multi-stage refinement. Combining these designs together forms an effective and efficient method that exploits all benefits of optimization-based and forward-based methods. Quantitative and qualitative results show that our E2Style performs better than existing forward-based methods and comparably to state-of-the-art optimization-based methods while maintaining the high efficiency as well as forward-based methods. Moreover, a number of real image editing applications demonstrate the efficacy of our E2Style. Our code is available at https://github.com/wty-ustc/e2style.

Recent research shows deep neural networks are vulnerable to different types of attacks, such as adversarial attacks, data poisoning attacks, and backdoor attacks. Among them, backdoor attacks are the most cunning and can occur in almost every stage of the deep learning pipeline. Backdoor attacks have attracted lots of interest from both academia and industry. However, most existing backdoor attack methods are visible or fragile to some effortless pre-processing such as common data transformations. To address these limitations, we propose a robust and invisible backdoor attack called "Poison Ink". Concretely, we first leverage the image structures as target poisoning areas and fill them with poison ink (information) to generate the trigger pattern. As the image structure can keep its semantic meaning during the data transformation, such a trigger pattern is inherently robust to data transformations. Then we leverage a deep injection network to embed such input-aware trigger pattern into the cover image to achieve stealthiness. Compared to existing popular backdoor attack methods, Poison Ink outperforms both in stealthiness and robustness. Through extensive experiments, we demonstrate that Poison Ink is not only general to different datasets and network architectures but also flexible for different attack scenarios. Besides, it also has very strong resistance against many state-of-the-art defense techniques.

With the rapid development of digital devices, the issue of how to transmit information among different devices with multimedia carriers has drawn much attention from the research community. This paper focuses on the important user scenario of “screen-to-camera information transmission”. Along this direction, image coding-based techniques have been shown to be the most popular and effective methods in the past decades. However, after careful study, we find that none of the existing methods can satisfy the four important properties simultaneously, i.e.,high transparency,high embedding efficiency,strong transmission robustnessandhigh adaptability to device types. This is mainly because these properties are contradictory with each other. In this paper, we thus propose a screen-to-camera image code dubbed “TERA” (transparency,efficiency,robustness andadaptability), which makes it possible to circumvent the contradiction among the above four properties for the first time. Generally, TERA adopts the color decomposition principle to ensure the visual quality and the superposition-based scheme to ensure embedding efficiency. BCH-coding-based information arrangement and a powerful attention-guided information decoding network are further designed to guarantee the robustness and adaptability. Through extensive experiments, the superiority and broad applications of our method are demonstrated.

Invertible grayscale is a special kind of grayscale from which the original color can be recovered. Given an input color image, this seminal work tries to hide the color information into its grayscale counterpart while making it hard to recognize any anomalies. This powerful functionality is enabled by training a hiding sub-network and restoring sub-network in an end-to-end way. Despite its expressive results, two key limitations exist: 1) The restored color image often suffers from some noticeable visual artifacts in the smooth regions. 2) It is very sensitive to JPEG compression, i.e., the original color information cannot be well recovered once the intermediate grayscale image is compressed by JPEG. To overcome these two limitations, this article introduces adversarial training and JPEG simulator respectively. Specifically, two auxiliary adversarial networks are incorporated to make the intermediate grayscale images and final restored color images indistinguishable from normal grayscale and color images. And the JPEG simulator is utilized to simulate real JPEG compression during the online training so that the hiding and restoring sub-networks can automatically learn to be JPEG robust. Extensive experiments demonstrate that the proposed method is superior to the original invertible grayscale work both qualitatively and quantitatively while ensuring the JPEG robustness. We further show that the proposed framework can be applied under different types of grayscale constraints and achieve excellent results.

Point cloud upsampling is vital for the quality of the mesh in three-dimensional reconstruction. Recent research on point cloud upsampling has achieved great success due to the development of deep learning. However, the existing methods regard point cloud upsampling of different scale factors as independent tasks. Thus, the methods need to train a specific model for each scale factor, which is both inefficient and impractical for storage and computation in real applications. To address this limitation, in this article, we propose a novel method called "Meta-PU" to first support point cloud upsampling of arbitrary scale factors with a single model. In the Meta-PU method, besides the backbone network consisting of residual graph convolution (RGC) blocks, a meta-subnetwork is learned to adjust the weights of the RGC blocks dynamically, and a farthest sampling block is adopted to sample different numbers of points. Together, these two blocks enable our Meta-PU to continuously upsample the point cloud with arbitrary scale factors by using only a single model. In addition, the experiments reveal that training on multiple scales simultaneously is beneficial to each other. Thus, Meta-PU even outperforms the existing methods trained for a specific scale factor only.

The complex nature of combining localization and classification in object detection has resulted in the flourished development of methods. Previous works tried to improve the performance in various object detection heads but failed to present a unified view. In this paper, we present a novel dynamic head framework to unify object detection heads with attentions. By coherently combining multiple self-attention mechanisms between feature levels for scale-awareness, among spatial locations for spatial-awareness, and within output channels for task-awareness, the proposed approach significantly improves the representation ability of object detection heads without any computational overhead. Further experiments demonstrate that the effectiveness and efficiency of the proposed dynamic head on the COCO benchmark. With a standard ResNeXt-101-DCN backbone, we largely improve the performance over popular object detectors and achieve a new state-of-the-art at 54.0 AP. The code will be released at https://github.com/microsoft/DynamicHead.

In this paper, we present a large scale unlabeled person re-identification (Re-ID) dataset "LUPerson" and make the first attempt of performing unsupervised pre-training for improving the generalization ability of the learned person Re-ID feature representation. This is to address the problem that all existing person Re-ID datasets are all of limited scale due to the costly effort required for data annotation. Previous research tries to leverage models pre-trained on ImageNet to mitigate the shortage of person Re-ID data but suffers from the large domain gap between ImageNet and person Re-ID data. LUPerson is an unlabeled dataset of 4M images of over 200K identities, which is 30× larger than the largest existing Re-ID dataset. It also covers a much diverse range of capturing environments (e.g., camera settings, scenes, etc.). Based on this dataset, we systematically study the key factors for learning Re-ID features from two perspectives: data augmentation and contrastive loss. Unsupervised pre-training performed on this large-scale dataset effectively leads to a generic Re-ID feature that can benefit all existing person Re-ID methods. Using our pre-trained model in some basic frameworks, our methods achieve state-of-the-art results without bells and whistles on four widely used Re-ID datasets: CUHK03, Market1501, DukeMTMC, and MSMT17. Our results also show that the performance improvement is more significant on small-scale target datasets or under few-shot setting.

Semantic image synthesis, translating semantic layouts to photo-realistic images, is a one-to-many mapping problem. Though impressive progress has been recently made, diverse semantic synthesis that can efficiently produce semantic-level multimodal results, still remains a challenge. In this paper, we propose a novel diverse semantic image synthesis framework from the perspective of semantic class distributions, which naturally supports diverse generation at semantic or even instance level. We achieve this by modeling class-level conditional modulation parameters as continuous probability distributions instead of discrete values, and sampling per-instance modulation parameters through instance-adaptive stochastic sampling that is consistent across the network. Moreover, we propose prior noise remapping, through linear perturbation parameters encoded from paired references, to facilitate supervised training and exemplar-based instance style control at test time. Extensive experiments on multiple datasets show that our method can achieve superior diversity and comparable quality compared to state-of-the-art methods. Code will be available at https://github.com/tzt101/INADE.git

Image matting is a fundamental and challenging problem in computer vision and graphics. Most existing matting methods leverage a user-supplied trimap as an auxiliary input to produce good alpha matte. However, obtaining high-quality trimap itself is arduous, thus restricting the application of these methods. Recently, some trimap-free methods have emerged, however, the matting quality is still far behind the trimap-based methods. The main reason is that, without the trimap guidance in some cases, the target network is ambiguous about which is the foreground target. In fact, choosing the foreground is a subjective procedure and depends on the user’s intention. To this end, this paper proposes an improved deep image matting framework which is trimap-free and only needs several user click interactions to eliminate the ambiguity. Moreover, we introduce a new uncertainty estimation module that can predict which parts need polishing and a following local refinement module. Based on the computation budget, users can choose how many local parts to improve with the uncertainty guidance. Quantitative and qualitative results show that our method performs better than existing trimap-free methods and comparably to state-of-the-art trimap-based methods with minimal user effort.

Face forgery by deepfake is widely spread over the internet and has raised severe societal concerns. Recently, how to detect such forgery contents has become a hot research topic and many deepfake detection methods have been proposed. Most of them model deepfake detection as a vanilla binary classification problem, i.e, first use a backbone network to extract a global feature and then feed it into a binary classifier (real/fake). But since the difference between the real and fake images in this task is often subtle and local, we argue this vanilla solution is not optimal. In this paper, we instead formulate deepfake detection as a fine-grained classification problem and propose a new multi-attentional deepfake detection network. Specifically, it consists of three key components: 1) multiple spatial attention heads to make the network attend to different local parts; 2) textural feature enhancement block to zoom in the subtle artifacts in shallow features; 3) aggregate the low-level textural feature and high-level semantic features guided by the attention maps. Moreover, to address the learning difficulty of this network, we further introduce a new regional independence loss and an attention guided data augmentation strategy. Through extensive experiments on different datasets, we demonstrate the superiority of our method over the vanilla binary classifier counterparts, and achieve state-of-the-art performance. The models will be released recently at https://github.com/yoctta/multiple-attention.

Unsupervised pretraining has achieved great success and many recent works have shown unsupervised pretraining can achieve comparable or even slightly better transfer performance than supervised pretraining on downstream target datasets. But in this paper, we find this conclusion may not hold when the target dataset has very few labeled samples for finetuning, i.e., few-label transfer. We analyze the possible reason from the clustering perspective: 1) The clustering quality of target samples is of great importance to few-label transfer; 2) Though contrastive learning is essential to learn how to cluster, its clustering quality is still inferior to supervised pretraining due to lack of label supervision. Based on the analysis, we interestingly discover that only involving some unlabeled target domain into the unsupervised pretraining can improve the clustering quality, subsequently reducing the transfer performance gap with supervised pretraining. This finding also motivates us to propose a new progressive few-label transfer algorithm for real applications, which aims to maximize the transfer performance under a limited annotation budget. To support our analysis and proposed method, we conduct extensive experiments on nine different target datasets. Experimental results show our proposed method can significantly boost the few-label transfer performance of unsupervised pretraining.

This paper aims at addressing the problem of substantial performance degradation at extremely low computational cost (e.g. 5M FLOPs on ImageNet classification). We found that two factors, sparse connectivity and dynamic activation function, are effective to improve the accuracy. The former avoids the significant reduction of network width, while the latter mitigates the detriment of reduction in network depth. Technically, we propose micro-factorized convolution, which factorizes a convolution matrix into low rank matrices, to integrate sparse connectivity into convolution. We also present a new dynamic activation function, named Dynamic Shift Max, to improve the non-linearity via maxing out multiple dynamic fusions between an input feature map and its circular channel shift. Building upon these two new operators, we arrive at a family of networks, named MicroNet, that achieves significant performance gains over the state of the art in the low FLOP regime. For instance, under the constraint of 12M FLOPs, MicroNet achieves 59.4% top-1 accuracy on ImageNet classification, outperforming MobileNetV3 by 9.6%. Source code is at https://github.com/liyunsheng13/micronet.

Image completion has made tremendous progress with convolutional neural networks (CNNs), because of their powerful texture modeling capacity. However, due to some inherent properties (e.g., local inductive prior, spatial-invariant kernels), CNNs do not perform well in understanding global structures or naturally support pluralistic completion. Recently, transformers demonstrate their power in modeling the long-term relationship and generating diverse results, but their computation complexity is quadratic to input length, thus hampering the application in processing high-resolution images. This paper brings the best of both worlds to pluralistic image completion: appearance prior reconstruction with transformer and texture replenishment with CNN. The former transformer recovers pluralistic coherent structures together with some coarse textures, while the latter CNN enhances the local texture details of coarse priors guided by the high-resolution masked images. The proposed method vastly outperforms state-of-the-art methods in terms of three aspects: 1) large performance boost on image fidelity even compared to deterministic completion methods; 2) better diversity and higher fidelity for pluralistic completion; 3) exceptional generalization ability on large masks and generic dataset, like ImageNet. Code and pre-trained models have been publicly released at https://github.com/raywzy/ICT.

Point cloud segmentation is a fundamental task in 3D. Despite recent progress on point cloud segmentation with the power of deep networks, current deep learning methods based on the clean label assumptions may fail with noisy labels. Yet, object class labels are often mislabeled in real-world point cloud datasets. In this work, we take the lead in solving this issue by proposing a novel Point Noise-Adaptive Learning (PNAL) framework. Compared to existing noise-robust methods on image tasks, our PNAL is noise-rate blind, to cope with the spatially variant noise rate problem specific to point clouds . Specifically, we propose a novel point-wise confidence selection to obtain reliable labels based on the historical predictions of each point. A novel cluster-wise label correction is proposed with a voting strategy to generate the best possible label taking the neighbor point correlations into consideration. We conduct extensive experiments to demonstrate the effectiveness of PNAL on both synthetic and real-world noisy datasets. In particular, even with 60% symmetric noisy labels, our proposed method produces much better results than its baseline counterpart without PNAL and is comparable to the ideal upper bound trained on a completely clean dataset. Moreover, we fully re-labeled the validation set of a popular but noisy real-world scene dataset ScanNetV2 to make it clean, for rigorous experiment and future research. Our code and data will be released.

Neural Architecture Search (NAS) often trains and evaluates a large number of architectures. Recent predictor-based NAS approaches attempt to alleviate such heavy computation costs with two key steps: sampling some architecture-performance pairs and fitting a proxy accuracy predictor. Given limited samples, these predictors, however, are far from accurate to locate top architectures due to the difficulty of fitting the huge search space. This paper reflects on a simple yet crucial question: if our final goal is to find the best architecture, do we really need to model the whole space well?. We propose a paradigm shift from fitting the whole architecture space using one strong predictor, to progressively fitting a search path towards the high-performance sub-space through a set of weaker predictors. As a key property of the weak predictors, their probabilities of sampling better architectures keep increasing. Hence we only sample a few well-performed architectures guided by the previously learned predictor and estimate a new better weak predictor. This embarrassingly easy framework, dubbed WeakNAS, produces coarse-to-fine iteration to gradually refine the ranking of sampling space. Extensive experiments demonstrate that WeakNAS costs fewer samples to find top-performance architectures on NAS-Bench-101 and NAS-Bench-201. Compared to state-of-the-art (SOTA) predictor-based NAS methods, WeakNAS outperforms all with notable margins, e.g., requiring at least 7.5x less samples to find global optimal on NAS-Bench-101. WeakNAS can also absorb their ideas to boost performance more. Further, WeakNAS strikes the new SOTA result of 81.3% in the ImageNet MobileNet Search Space. The code is available at: https://github.com/VITA-Group/WeakNAS.

Image style transfer is to re-render the content of one image with the style of another. Most existing methods couple content and style information in their network structures and hyper-parameters, and learn it as a black-box. For better understanding, this paper aims to provide a new explicit decoupled perspective. Specifically, we propose StyleBank, which is composed of multiple convolution filter banks and each filter bank explicitly represents one style. To transfer an image to a specific style, the corresponding filter bank is operated on the intermediate feature produced by a single auto-encoder. The StyleBank and the auto-encoder are jointly learnt in such a way that the auto-encoder does not encode any style information. This explicit representation also enables us to conduct incremental learning to add a new style and fuse styles at not only the image level, but also the region level. Our method is the first style transfer network that links back to traditional texton mapping methods, and provides new understanding on neural style transfer. We further apply this general filterbank learning idea to two different multi-parameter image processing tasks: edge-aware image smoothing and denoising. Experiments demonstrate that it can achieve comparable results to its single parameter setting counterparts.

Many different deep networks have been used to approximate, accelerate or improve traditional image operators. Among these traditional operators, many contain parameters which need to be tweaked to obtain the satisfactory results, which we refer to as "parameterized image operators". However, most existing deep networks trained for these operators are only designed for one specific parameter configuration, which does not meet the needs of real scenarios that usually require flexible parameters settings. To overcome this limitation, we propose a new decoupled learning algorithm to learn from the operator parameters to dynamically adjust the weights of a deep network for image operators, denoted as the base network. The learned algorithm is formed as another network, namely the weight learning network, which can be end-to-end jointly trained with the base network. Experiments demonstrate that the proposed framework can be successfully applied to many traditional parameterized image operators. To accelerate the parameter tuning for practical scenarios, the proposed framework can be further extended to dynamically change the weights of only one single layer of the base network while sharing most computation cost. We demonstrate that this cheap parameter-tuning extension of the proposed decoupled learning framework even outperforms the state-of-the-art alternative approaches.

Traditional watermarking algorithms have been extensively studied. As an important type of watermarking schemes, template-based approaches maintain a very high embedding rate. In such scheme, the message is often represented by some dedicatedly designed templates, and then the message embedding process is carried out by additive operation with the templates and the host image. To resist potential distortions, these templates often need to contain some special statistical features so that they can be successfully recovered at the extracting side. But in existing methods, most of these features are handcrafted and too simple, thus making them not robust enough to resist serious distortions unless very strong and obvious templates are used. Inspired by the powerful feature learning capacity of deep neural network, we propose the first deep template-based watermarking algorithm in this paper. Specifically, at the embedding side, we first design two new templates for message embedding and locating, which is achieved by leveraging the special properties of human visual system, i.e., insensitivity to specific chrominance components, the proximity principle and the oblique effect. At the extracting side, we propose a novel two-stage deep neural network, which consists of an auxiliary enhancing sub-network and a classification sub-network. Thanks to the power of deep neural networks, our method achieves both digital editing resilience and camera shooting resilience based on typical application scenarios. Through extensive experiments, we demonstrate that the proposed method can achieve much better robustness than existing methods while guaranteeing the original visual quality.

Deep learning has achieved tremendous success in numerous industrial applications. As training a good model often needs massive high-quality data and computation resources, the learned models often have significant business values. However, these valuable deep models are exposed to a huge risk of infringements. For example, if the attacker has the full information of one target model including the network structure and weights, the model can be easily finetuned on new datasets. Even if the attacker can only access the output of the target model, he/she can still train another similar surrogate model by generating a large scale of input-output training pairs. How to protect the intellectual property of deep models is a very important but seriously under-researched problem. There are a few recent attempts at classification network protection only.In this paper, we propose the first model watermarking framework for protecting image processing models. To achieve this goal, we leverage the spatial invisible watermarking mechanism. Specifically, given a black-box target model, a unified and invisible watermark is hidden into its outputs, which can be regarded as a special task-agnostic barrier. In this way, when the attacker trains one surrogate model by using the input-output pairs of the target model, the hidden watermark will be learned and extracted afterward. To enable watermarks from binary bits to high-resolution images, both traditional and deep spatial invisible watermarking mechanism are considered. Experiments demonstrate the robustness of the proposed watermarking mechanism, which can resist surrogate models learned with different network structures and objective functions. Besides deep models, the proposed method is also easy to be extended to protect data and traditional image processing algorithms.

Light-weight convolutional neural networks (CNNs) suffer performance degradation as their low computational budgets constrain both the depth (number of convolution layers) and the width (number of channels) of CNNs, resulting in limited representation capability. To address this issue, we present Dynamic Convolution, a new design that increases model complexity without increasing the network depth or width. Instead of using a single convolution kernel per layer, dynamic convolution aggregates multiple parallel convolution kernels dynamically based upon their attentions, which are input dependent. Assembling multiple kernels is not only computationally efficient due to the small kernel size, but also has more representation power since these kernels are aggregated in a non-linear way via attention. By simply using dynamic convolution for the state-of-the-art architecture MobileNetV3-Small, the top-1 accuracy of ImageNet classification is boosted by 2.9% with only 4% additional FLOPs and 2.9 AP gain is achieved on COCO keypoint detection.

In this paper, we look into the problem of 3D adversary attack, and propose to leverage the internal properties of the point clouds and the adversarial examples to design a new self-robust deep neural network (DNN) based 3D recognition systems. As a matter of fact, on one hand, point clouds are highly structured. Hence for each local part of clean point clouds, it is possible to learn what is it (``part of a bottle") and its relative position (``upper part of a bottle") to the global object center. On the other hand, with the visual quality constraint, 3D adversarial samples often only produce small local perturbations, thus they will roughly keep the original global center but may cause incorrect local relative position estimation. Motivated by these two properties, we use relative position (dubbed as ``gather-vector") as the adversarial indicator and propose a new robust gather module. Equipped with this module, we further propose a new self-robust 3D point recognition network. Through extensive experiments, we demonstrate that the proposed method can improve the robustness of the target attack under the white-box setting significantly. For I-FGSM based attack, our method reduces the attack success rate from 94.37 \% to 75.69 \%. For C\&W based attack, our method reduces the attack success rate more than 40.00 \%. Moreover, our method is complementary to other types of defense methods to achieve better defense results.

Deep Neural Networks are vulnerable to adversarial samples, which can fool classifiers by adding small perturbations onto the original image. Since the pioneering optimization-based adversarial attack method, many following methods have been proposed in the past several years. However most of these methods add perturbations in a "pixel-wise" and "global" way. Firstly, because of the contradiction between the local smoothness of natural images and the noisy property of these adversarial perturbations, this "pixel-wise" way makes these methods not robust to image processing based defense methods and steganalysis based detection methods. Secondly, we find adding perturbations to the background is less useful than to the salient object, thus the "global" way is also not optimal. Based on these two considerations, we propose the first robust superpixel-guided attentional adversarial attack method. Specifically, the adversarial perturbations are only added to the salient regions and guaranteed to be same within each superpixel. Through extensive experiments, we demonstrate our method can preserve the attack ability even in this highly constrained modification space. More importantly, compared to existing methods, it is significantly more robust to image processing based defense and steganalysis based detection.

Semi-supervised learning (SSL) has been extensively studied to improve the generalization ability of deep neural networks for visual recognition. To involve the unlabelled data, most existing SSL methods are based on common density-based cluster assumption: samples lying in the same high-density region are likely to belong to the same class, including the methods performing consistency regularization or generating pseudo-labels for the unlabelled images. Despite their impressive performance, we argue three limitations exist: 1) Though the density information is demonstrated to be an important clue, they all use it in an implicit way and have not exploited it in depth. 2) For feature learning, they often learn the feature embedding based on the single data sample and ignore the neighborhood information. 3) For label-propagation based pseudo-label generation, it is often done offline and difficult to be end-to-end trained with feature learning. Motivated by these limitations, this paper proposes to solve the SSL problem by building a novel density-aware graph, based on which the neighborhood information can be easily leveraged and the feature learning and label propagation can also be trained in an end-to-end way. Specifically, we first propose a new Density-aware Neighborhood Aggregation(DNA) module to learn more discriminative features by incorporating the neighborhood information in a density-aware manner. Then a novel Density-ascending Path based Label Propagation(DPLP) module is proposed to generate the pseudo-labels for unlabeled samples more efficiently according to the feature distribution characterized by density. Finally, the DNA module and DPLP module evolve and improve each other end-to-end. Extensive experiments demonstrate the effectiveness of the newly proposed density-aware graph based SSL framework and our approach can outperform current state-of-the-art methods by a large margin.

We propose to restore old photos that suffer from severe degradation through a deep learning approach. Unlike conventional restoration tasks that can be solved through supervised learning, the degradation in real photos is complex and the domain gap between synthetic images and real old photos makes the network fail to generalize. Therefore, we propose a novel triplet domain translation network by leveraging real photos along with massive synthetic image pairs. Specifically, we train two variational autoencoders (VAEs) to respectively transform old photos and clean photos into two latent spaces. And the translation between these two latent spaces is learned with synthetic paired data. This translation generalizes well to real photos because the domain gap is closed in the compact latent space. Besides, to address multiple degradations mixed in one old photo, we design a global branch with a partial nonlocal block targeting to the structured defects, such as scratches and dust spots, and a local branch targeting to the unstructured defects, such as noises and blurriness. Two branches are fused in the latent space, leading to improved capability to restore old photos from multiple defects. The proposed method outperforms state-of-the-art methods in terms of visual quality for old photos restoration.

Deep neural networks have made tremendous progress in 3D point-cloud recognition. Recent works have shown that these 3D recognition networks are also vulnerable to adversarial samples produced from various attack methods, including optimization-based 3D Carlini-Wagner attack, gradient-based iterative fast gradient method, and skeleton-detach based point-dropping. However, after a careful analysis, these methods are either extremely slow because of the optimization/iterative scheme, or not flexible to support targeted attack of a specific category. To overcome these shortcomings, this paper proposes a novel label guided adversarial network (LG-GAN) for real-time flexible targeted point cloud attack. To the best of our knowledge, this is the first generation based 3D point cloud attack method. By feeding the original point clouds and target attack label into LG-GAN, it can learn how to deform the point clouds to mislead the recognition network into the specific label only with a single forward pass. In detail, LG-GAN first leverages one multi-branch adversarial network to extract hierarchical features of the input point clouds, then incorporates the specified label information into multiple intermediate features using the label encoder. Finally, the encoded features will be fed into the coordinate reconstruction decoder to generate the target adversarial sample. By evaluating different point-cloud recognition models (e.g., PointNet, PointNet++ and DGCNN), we demonstrate that the proposed LG-GAN can support flexible targeted attack on the fly while guaranteeing good attack performance and higher efficiency simultaneously.

Modern deep neural networks(DNNs) are vulnerable to adversarial samples. Sparse adversarial samples are a special branch of adversarial samples that can fool the target model by only perturbing a few pixels. The existence of the sparse adversarial attack points out that DNNs are much more vulnerable than people believed, which is also a new aspect for analyzing DNNs. However, current sparse adversarial attack methods still have some shortcomings on both sparsity and invisibility. In this paper, we propose a novel two-stage distortion-aware greedy-based method dubbed as ''GreedyFool". Specifically, it first selects the most effective candidate positions to modify by considering both the gradient(for adversary) and the distortion map(for invisibility), then drops some less important points in the reduce stage. Experiments demonstrate that compared with the start-of-the-art method, we only need to modify 3 times fewer pixels under the same sparse perturbation setting. For target attack, the success rate of our method is 9.96% higher than the start-of-the-art method under the same pixel budget.

Despite tremendous success in many application scenarios, deep learning faces serious intellectual property (IP) infringement threats. Considering the cost of designing and training a good model, infringements will significantly infringe the interests of the original model owner. Recently, many impressive works have emerged for deep model IP protection. However, they either are vulnerable to ambiguity attacks, or require changes in the target network structure by replacing its original normalization layers and hence cause significant performance drops. To this end, we propose a new passport-aware normalization formulation, which is generally applicable to most existing normalization layers and only needs to add another passport-aware branch for IP protection. This new branch is jointly trained with the target model but discarded in the inference stage. Therefore it causes no structure change in the target model. Only when the model IP is suspected to be stolen by someone, the private passport-aware branch is added back for ownership verification. Through extensive experiments, we verify its effectiveness in both image and 3D point recognition models. It is demonstrated to be robust not only to common attack techniques like fine-tuning and model compression, but also to ambiguity attacks. By further combining it with trigger-set based methods, both black-box and white-box verification can be achieved for enhanced security of deep learning models deployed in real systems.

Traditional image processing operators often provide some control parameters to tweak the final results. Recently, different convolutional neural networks have been used to approximate or improve these operators. However, in those methods, one single model can only handle one operator of a specific parameter value and does not support parameter tuning. In this paper, we propose a new plugin module, “Adaptive Filterbank Pyramid”, which can be inserted into a backbone network to support multiple operators and continuous parameter tuning. Our module explicitly represents one operator with one filterbank pyramid. To generate the results of a specific operator, the corresponding filterbank pyramid is convolved with the intermediate feature pyramid produced by the backbone network. The weights of the filterbank pyramid are directly regressed by another sub-network, which is jointly trained with the backbone network and adapted to the input parameter, thus enabling continuous parameter tuning. We applied the proposed module for a large variety of image processing tasks, including image smoothing, image denoising, image deblocking, image enhancement and neural style transfer. Experiments show that our method is generalized to different types of image processing tasks and different backbone network structures. Compared to the single-operator-single-parameter baseline, our method can produce comparable results but is significantly more efficient in both training and testing.

Our impression about one person often updates after we see more aspects of him/her and this process keeps iterating given more meetings. We formulate such an intuition into the problem of person re-identification (re-ID), where the representation of a query (probe) image is iteratively updated with new information from the candidates in the gallery. Specifically, we propose a simple attentional aggregation formulation to instantiate this idea and showcase that such a pipeline achieves competitive performance on standard benchmarks including CUHK03, Market-1501 and DukeMTMC. Not only does such a simple method improve the performance of the baseline models, it also achieves comparable performance with latest advanced re-ranking methods. Another advantage of this proposal is its flexibility to incorporate different representations and similarity metrics. By utilizing stronger representations and metrics, we further demonstrate state-of-the-art person re-ID performance, which also validates the general applicability of the proposed method.

Despite the recent success of face image generation with GANs, conditional hair editing remains challenging due to the under-explored complexity of its geometry and appearance. In this paper, we present MichiGAN (Multi-Input-Conditioned Hair Image GAN), a novel conditional image generation method for interactive portrait hair manipulation. To provide user control over every major hair visual factor, we explicitly disentangle hair into four orthogonal attributes, including shape, structure, appearance, and background. For each of them, we design a corresponding condition module to represent, process, and convert user inputs, and modulate the image generation pipeline in ways that respect the natures of different visual attributes. All these condition modules are integrated with the backbone generator to form the final end-to-end network, which allows fully-conditioned hair generation from multiple user inputs. Upon it, we also build an interactive portrait hair editing system that enables straightforward manipulation of hair by projecting intuitive and high-level user inputs such as painted masks, guiding strokes, or reference photos to well-defined condition representations. Through extensive experiments and evaluations, we demonstrate the superiority of our method regarding both result quality and user controllability.

Modern deep neural networks are often vulnerable to adversarial samples. Based on the first optimization-based attacking method, many following methods are proposed to improve the attacking performance and speed. Recently, generation-based methods have received much attention since they directly use feed-forward networks to generate the adversarial samples, which avoid the time-consuming iterative attacking procedure in optimization-based and gradient-based methods. However, current generation-based methods are only able to attack one specific target (category) within one model, thus making them not applicable to real classification systems that often have hundreds/thousands of categories. In this paper, we propose the first Multi-target Adversarial Network (MAN), which can generate multi-target adversarial samples with a single model. By incorporating the specified category information into the intermediate features, it can attack any category of the target classification model during runtime. Experiments show that the proposed MAN can produce stronger attack results and also have better transferability than previous state-of-the-art methods in both multi-target attack task and single-target attack task. We further use the adversarial samples generated by our MAN to improve the robustness of the classification model. It can also achieve better classification accuracy than other methods when attacked by various methods.

To recognize objects of the unseen classes, most existing Zero-Shot Learning (ZSL) methods first learn a compatible projection function between the common semantic space and the visual space based on the data of source seen classes, then directly apply it to the target unseen classes. However, in real scenarios, the data distribution between the source and target domain might not match well, thus causing the well-known domain shift problem. Based on the observation that visual features of test instances can be separated into different clusters, we propose a new visual structure constraint on class centers for transductive ZSL, to improve the generality of the projection function (\ie alleviate the above domain shift problem). Specifically, three different strategies (symmetric Chamfer-distance,Bipartite matching distance, and Wasserstein distance) are adopted to align the projected unseen semantic centers and visual cluster centers of test instances. We also propose a new training strategy to handle the real cases where many unrelated images exist in the test dataset, which is not considered in previous methods. Experiments on many widely used datasets demonstrate that the proposed visual structure constraint can bring substantial performance gain consistently and achieve state-of-the-art results.

Image dehazing aims to recover the uncorrupted content from a hazy image. Instead of leveraging traditional low-level or handcrafted image priors as the restoration constraints, e.g., dark channels and increased contrast, we propose an end-to-end gated context aggregation network to directly restore the final haze-free image. In this network, we adopt the latest smoothed dilation technique to help remove the gridding artifacts caused by the widely-used dilated convolution with negligible extra parameters, and leverage a gated sub-network to fuse the features from different levels. Extensive experiments demonstrate that our method can surpass previous state-of-the-art methods by a large margin both quantitatively and qualitatively. In addition, to demonstrate the generality of the proposed method, we further apply it to the image deraining task, which also achieves the state-of-the-art performance.

Removing reflection artefacts from a single image is a problem of both theoretical and practical interest, which still presents challenges because of the massively ill-posed nature of the problem. In this paper, we propose a technique based on a novel optimization problem. First, we introduce a simple user interaction scheme, which helps minimize information loss in the reflection-free regions. Second, we introduce an H2fidelity term, which preserves fine detail while enforcing the global color similarity. We show that this combination allows us to mitigate the shortcomings in structure and color preservation, which presents some of the most prominent drawbacks in the existing methods for reflection removal. We demonstrate, through numerical and visual experiments, that our method is able to outperform the state-of-the-art model-based methods and compete with recent deep-learning approaches.

We propose a new algorithm for color transfer between images that have perceptually similar semantic structures. We aim to achieve a more accurate color transfer that leverages semantically meaningful dense correspondence between images. To accomplish this, our algorithm uses neural representations for matching. Additionally, the color transfer should be spatially variant and globally coherent. Therefore, our algorithm optimizes a local linear model for color transfer satisfying both local and global constraints. Our proposed approach jointly optimizes matching and color transfer, adopting a coarse-to-fine strategy. The proposed method can be successfully extended from one-to-one to one-to-many color transfer. The latter further addresses the problem of mismatching elements of the input image. We validate our proposed method by testing it on a large variety of image content.

This paper presents the first attempt at stereoscopic neural style transfer, which responds to the emerging demand for 3D movies or AR/VR. We start with a careful examination of applying existing monocular style transfer methods to left and right views of stereoscopic images separately. This reveals that the original disparity consistency cannot be well preserved in the final stylization results, which causes 3D fatigue to the viewers. To address this issue, we incorporate a new disparity loss into the widely adopted style loss function by enforcing the bidirectional disparity constraint in non-occluded regions. For a practical realtime solution, we propose the first feed-forward network by jointly training a stylization sub-network and a disparity sub-network, and integrate them in a feature level middle domain. Our disparity sub-network is also the first end-to-end network for simultaneous bidirectional disparity and occlusion mask estimation. Finally, our network is effectively extended to stereoscopic videos, by considering both temporal coherence and disparity consistency. We will show that the proposed method clearly outperforms the baseline algorithms both quantitatively and qualitatively.

We propose the first deep learning approach for exemplar-based local colorization. Given a reference color image, our convolutional neural network directly maps a grayscale image to an output colorized image. Rather than using hand-crafted rules as in traditional exemplar-based methods, our end-to-end colorization network learns how to select, propagate , and predict colors from the large-scale data. The approach performs robustly and generalizes well even when using reference images that are unrelated to the input grayscale image. More importantly, as opposed to other learning-based colorization methods, our network allows the user to achieve customizable results by simply feeding different references. In order to further reduce manual effort in selecting the references, the system automatically recommends references with our proposed image retrieval algorithm, which considers both semantic and luminance information. The colorization can be performed fully automatically by simply picking the top reference suggestion. Our approach is validated through a user study and favorable quantitative comparisons to the-state-of-the-art methods. Furthermore, our approach can be naturally extended to video colorization. Our code and models are freely available for public use.

We propose StyleBank, which is composed of multiple convolution filter banks and each filter bank explicitly represents one style, for neural image style transfer. To transfer an image to a specific style, the corresponding filter bank is operated on top of the intermediate feature embedding produced by a single auto-encoder. The StyleBank and the auto-encoder are jointly learnt, where the learning is conducted in such a way that the auto-encoder does not encode any style information thanks to the flexibility introduced by the explicit filter bank representation. It also enables us to conduct incremental learning to add a new image style by learning a new filter bank while holding the auto-encoder fixed. The explicit style representation along with the flexible network design enables us to fuse styles at not only the image level, but also the region level. Our method is the first style transfer network that links back to traditional texton mapping methods, and hence provides new understanding on neural style transfer. Our method is easy to train, runs in real-time, and produces results that qualitatively better or at least comparable to existing methods.

Training a feed-forward network for the fast neural style transfer of images has proven successful, but the naive extension of processing videos frame by frame is prone to producing flickering results. We propose the first end-to-end network for online video style transfer, which generates temporally coherent stylized video sequences in near realtime. Two key ideas include an efficient network by incorporating short-term coherence, and propagating short-term coherence to long-term, which ensures consistency over a longer period of time. Our network can incorporate different image stylization networks and clearly outperforms the per-frame baseline both qualitatively and quantitatively. Moreover, it can achieve visually comparable coherence to optimization-based video style transfer, but is three orders of magnitude faster.