Yanye Lu

Lei Zhu, Qi She, Qian Chen et al.

Weakly supervised object localization (WSOL) focuses on localizing objects only with the supervision of image-level classification masks. Most previous WSOL methods follow the classification activation map (CAM) that localizes objects based on the classification structure with the multi-instance learning (MIL) mechanism. However, the MIL mechanism makes CAM only activate discriminative object parts rather than the whole object, weakening its performance for localizing objects. To avoid this problem, this work provides a novel perspective that models WSOL as a domain adaption (DA) task, where the score estimator trained on the source/image domain is tested on the target/pixel domain to locate objects. Under this perspective, a DA-WSOL pipeline is designed to better engage DA approaches into WSOL to enhance localization performance. It utilizes a proposed target sampling strategy to select different types of target samples. Based on these types of target samples, domain adaption localization (DAL) loss is elaborated. It aligns the feature distribution between the two domains by DA and makes the estimator perceive target domain cues by Universum regularization. Experiments show that our pipeline outperforms SOTA methods on multi benchmarks. Code are released at \url{https://github.com/zh460045050/DA-WSOL_CVPR2022}.

Lei Zhu, Qian Chen, Lujia Jin et al.

Classification activation map (CAM), utilizing the classification structure to generate pixel-wise localization maps, is a crucial mechanism for weakly supervised object localization (WSOL). However, CAM directly uses the classifier trained on image-level features to locate objects, making it prefers to discern global discriminative factors rather than regional object cues. Thus only the discriminative locations are activated when feeding pixel-level features into this classifier. To solve this issue, this paper elaborates a plug-and-play mechanism called BagCAMs to better project a well-trained classifier for the localization task without refining or re-training the baseline structure. Our BagCAMs adopts a proposed regional localizer generation (RLG) strategy to define a set of regional localizers and then derive them from a well-trained classifier. These regional localizers can be viewed as the base learner that only discerns region-wise object factors for localization tasks, and their results can be effectively weighted by our BagCAMs to form the final localization map. Experiments indicate that adopting our proposed BagCAMs can improve the performance of baseline WSOL methods to a great extent and obtains state-of-the-art performance on three WSOL benchmarks. Code are released at https://github.com/zh460045050/BagCAMs.

Shuang Zeng, Lei Zhu, Xinliang Zhang et al. · pku

Medical image segmentation is a fundamental yet challenging task due to the arduous process of acquiring large volumes of high-quality labeled data from experts. Contrastive learning offers a promising but still problematic solution to this dilemma. Firstly existing medical contrastive learning strategies focus on extracting image-level representation, which ignores abundant multi-level representations. Furthermore they underutilize the decoder either by random initialization or separate pre-training from the encoder, thereby neglecting the potential collaboration between the encoder and decoder. To address these issues, we propose a novel multi-level asymmetric contrastive learning framework named MACL for volumetric medical image segmentation pre-training. Specifically, we design an asymmetric contrastive learning structure to pre-train encoder and decoder simultaneously to provide better initialization for segmentation models. Moreover, we develop a multi-level contrastive learning strategy that integrates correspondences across feature-level, image-level, and pixel-level representations to ensure the encoder and decoder capture comprehensive details from representations of varying scales and granularities during the pre-training phase. Finally, experiments on 8 medical image datasets indicate our MACL framework outperforms existing 11 contrastive learning strategies. i.e. Our MACL achieves a superior performance with more precise predictions from visualization figures and 1.72%, 7.87%, 2.49% and 1.48% Dice higher than previous best results on ACDC, MMWHS, HVSMR and CHAOS with 10% labeled data, respectively. And our MACL also has a strong generalization ability among 5 variant U-Net backbones. Our code will be released at https://github.com/stevezs315/MACL.

Lei Zhu, Hangzhou He, Xinliang Zhang et al. · pku

End-to-end weakly supervised semantic segmentation aims at optimizing a segmentation model in a single-stage training process based on only image annotations. Existing methods adopt an online-trained classification branch to provide pseudo annotations for supervising the segmentation branch. However, this strategy makes the classification branch dominate the whole concurrent training process, hindering these two branches from assisting each other. In our work, we treat these two branches equally by viewing them as diverse ways to generate the segmentation map, and add interactions on both their supervision and operation to achieve mutual promotion. For this purpose, a bidirectional supervision mechanism is elaborated to force the consistency between the outputs of these two branches. Thus, the segmentation branch can also give feedback to the classification branch to enhance the quality of localization seeds. Moreover, our method also designs interaction operations between these two branches to exchange their knowledge to assist each other. Experiments indicate our work outperforms existing end-to-end weakly supervised segmentation methods.

Lujia Jin, Shi Zhao, Lei Zhu et al.

The performance of learning-based denoising largely depends on clean supervision. However, it is difficult to obtain clean images in many scenes. On the contrary, the capture of multiple noisy frames for the same field of view is available and often natural in real life. Therefore, it is necessary to avoid the restriction of clean labels and make full use of noisy data for model training. So we propose an unsupervised learning strategy named one-pot denoising (OPD) for multi-frame images. OPD is the first proposed unsupervised multi-frame denoising (MFD) method. Different from the traditional supervision schemes including both supervised Noise2Clean (N2C) and unsupervised Noise2Noise (N2N), OPD executes mutual supervision among all of the multiple frames, which gives learning more diversity of supervision and allows models to mine deeper into the correlation among frames. N2N has also been proved to be actually a simplified case of the proposed OPD. From the perspectives of data allocation and loss function, two specific implementations, random coupling (RC) and alienation loss (AL), are respectively provided to accomplish OPD during model training. In practice, our experiments demonstrate that OPD behaves as the SOTA unsupervised denoising method and is comparable to supervised N2C methods for synthetic Gaussian and Poisson noise, and real-world optical coherence tomography (OCT) speckle noise.

Lei Zhu, Fangyun Wei, Yanye Lu

In this work, we investigate the potential of a large language model (LLM) to directly comprehend visual signals without the necessity of fine-tuning on multi-modal datasets. The foundational concept of our method views an image as a linguistic entity, and translates it to a set of discrete words derived from the LLM's vocabulary. To achieve this, we present the Vision-to-Language Tokenizer, abbreviated as V2T Tokenizer, which transforms an image into a ``foreign language'' with the combined aid of an encoder-decoder, the LLM vocabulary, and a CLIP model. With this innovative image encoding, the LLM gains the ability not only for visual comprehension but also for image denoising and restoration in an auto-regressive fashion-crucially, without any fine-tuning. We undertake rigorous experiments to validate our method, encompassing understanding tasks like image recognition, image captioning, and visual question answering, as well as image denoising tasks like inpainting, outpainting, deblurring, and shift restoration. Code and models are available at https://github.com/zh460045050/V2L-Tokenizer.

JiaKui Hu, Jialun Liu, Liying Yang et al.

Scene-consistent video generation aims to create videos that explore 3D scenes based on a camera trajectory. Previous methods rely on video generation models with external memory for consistency, or iterative 3D reconstruction and inpainting, which accumulate errors during inference due to incorrect intermediary outputs, non-differentiable processes, and separate models. To overcome these limitations, we introduce ``geometry-as-context". It iteratively completes the following steps using an autoregressive camera-controlled video generation model: (1) estimates the geometry of the current view necessary for 3D reconstruction, and (2) simulates and restores novel view images rendered by the 3D scene. Under this multi-task framework, we develop the camera gated attention module to enhance the model's capability to effectively leverage camera poses. During the training phase, text contexts are utilized to ascertain whether geometric or RGB images should be generated. To ensure that the model can generate RGB-only outputs during inference, the geometry context is randomly dropped from the interleaved text-image-geometry training sequence. The method has been tested on scene video generation with one-direction and forth-and-back trajectories. The results show its superiority over previous approaches in maintaining scene consistency and camera control.

Shuang Zeng, Boxu Xie, Lei Zhu et al.

Deep learning has greatly advanced medical image segmentation, but its success relies heavily on fully supervised learning, which requires dense annotations that are costly and time-consuming for 3D volumetric scans. Barely-supervised learning reduces annotation burden by using only a few labeled slices per volume. Existing methods typically propagate sparse annotations to unlabeled slices through geometric continuity to generate pseudo-labels, but this strategy lacks semantic understanding, often resulting in low-quality pseudo-labels. Furthermore, medical image segmentation is inherently a pixel-level visual understanding task, where accuracy fundamentally depends on the quality of local, fine-grained visual features. Inspired by this, we propose RADA, a novel Region-Aware Dual-encoder Auxiliary learning pipeline which introduces a dual-encoder framework pre-trained on Alpha-CLIP to extract fine-grained, region-specific visual features from the original images and limited annotations. The framework combines image-level fine-grained visual features with text-level semantic guidance, providing region-aware semantic supervision that bridges image-level semantics and pixel-level segmentation. Integrated into a triple-view training framework, RADA achieves SOTA performance under extremely sparse annotation settings on LA2018, KiTS19 and LiTS, demonstrating robust generalization across diverse datasets.

JiaKui Hu, Lujia Jin, Zhengjian Yao et al.

This paper proposes the Degradation Classification Pre-Training (DCPT), which enables models to learn how to classify the degradation type of input images for universal image restoration pre-training. Unlike the existing self-supervised pre-training methods, DCPT utilizes the degradation type of the input image as an extremely weak supervision, which can be effortlessly obtained, even intrinsic in all image restoration datasets. DCPT comprises two primary stages. Initially, image features are extracted from the encoder. Subsequently, a lightweight decoder, such as ResNet18, is leveraged to classify the degradation type of the input image solely based on the features extracted in the first stage, without utilizing the input image. The encoder is pre-trained with a straightforward yet potent DCPT, which is used to address universal image restoration and achieve outstanding performance. Following DCPT, both convolutional neural networks (CNNs) and transformers demonstrate performance improvements, with gains of up to 2.55 dB in the 10D all-in-one restoration task and 6.53 dB in the mixed degradation scenarios. Moreover, previous self-supervised pretraining methods, such as masked image modeling, discard the decoder after pre-training, while our DCPT utilizes the pre-trained parameters more effectively. This superiority arises from the degradation classifier acquired during DCPT, which facilitates transfer learning between models of identical architecture trained on diverse degradation types. Source code and models are available at https://github.com/MILab-PKU/dcpt.

Zhengjian Yao, Jiakui Hu, Kaiwen Li et al.

Blind face restoration remains a persistent challenge due to the inherent ill-posedness of reconstructing holistic structures from severely constrained observations. Current generative approaches, while capable of synthesizing realistic textures, often suffer from information asymmetry -- the intrinsic disparity between the information-sparse low quality inputs and the information-dense high quality outputs. This imbalance leads to a one-to-many mapping, where insufficient constraints result in stochastic uncertainty and hallucinatory artifacts. To bridge this gap, we present \textbf{Pref-Restore}, a hierarchical framework that integrates discrete semantic logic with continuous texture generation to achieve deterministic, preference-aligned restoration. Our methodology fundamentally addresses this information disparity through two complementary strategies: (1) Augmenting Input Density: We employ an auto-regressive integrator to reformulate textual instructions into dense latent queries, injecting high-level semantic stability to constrain the degraded signals; (2) Pruning Output Distribution: We pioneer the integration of on-policy reinforcement learning directly into the diffusion restoration loop. By transforming human preferences into differentiable constraints, we explicitly penalize stochastic deviations, thereby sharpening the posterior distribution toward the desired high-fidelity outcomes. Extensive experiments demonstrate that Pref-Restore achieves state-of-the-art performance across synthetic and real-world benchmarks. Furthermore, empirical analysis confirms that our preference-aligned strategy significantly reduces solution entropy, establishing a robust pathway toward reliable and deterministic blind restoration.

Xinliang Zhang, Lei Zhu, Hangzhou He et al. · pku

Scribble-based weakly-supervised semantic segmentation using sparse scribble supervision is gaining traction as it reduces annotation costs when compared to fully annotated alternatives. Existing methods primarily generate pseudo-labels by diffusing labeled pixels to unlabeled ones with local cues for supervision. However, this diffusion process fails to exploit global semantics and class-specific cues, which are important for semantic segmentation. In this study, we propose a class-driven scribble promotion network, which utilizes both scribble annotations and pseudo-labels informed by image-level classes and global semantics for supervision. Directly adopting pseudo-labels might misguide the segmentation model, thus we design a localization rectification module to correct foreground representations in the feature space. To further combine the advantages of both supervisions, we also introduce a distance entropy loss for uncertainty reduction, which adapts per-pixel confidence weights according to the reliable region determined by the scribble and pseudo-label's boundary. Experiments on the ScribbleSup dataset with different qualities of scribble annotations outperform all the previous methods, demonstrating the superiority and robustness of our method.The code is available at https://github.com/Zxl19990529/Class-driven-Scribble-Promotion-Network.

JiaKui Hu, Shanshan Zhao, Qing-Guo Chen et al.

This paper presents Omni-View, which extends the unified multimodal understanding and generation to 3D scenes based on multiview images, exploring the principle that "generation facilitates understanding". Consisting of understanding model, texture module, and geometry module, Omni-View jointly models scene understanding, novel view synthesis, and geometry estimation, enabling synergistic interaction between 3D scene understanding and generation tasks. By design, it leverages the spatiotemporal modeling capabilities of its texture module responsible for appearance synthesis, alongside the explicit geometric constraints provided by its dedicated geometry module, thereby enriching the model's holistic understanding of 3D scenes. Trained with a two-stage strategy, Omni-View achieves a state-of-the-art score of 55.4 on the VSI-Bench benchmark, outperforming existing specialized 3D understanding models, while simultaneously delivering strong performance in both novel view synthesis and 3D scene generation.

Qian Chen, Yuehao Chen, Qiang Wang et al.

Retinal laser speckle contrast imaging (LSCI) is a noninvasive optical modality for monitoring retinal blood flow dynamics. However, conventional temporal LSCI (tLSCI) reconstruction relies on sufficiently long speckle sequences to obtain stable temporal statistics, which makes it vulnerable to acquisition disturbances and limits effective temporal resolution. A physically informed reconstruction framework, termed RetinaDiff (Retinal Diffusion Model), is proposed for retinal tLSCI that is robust to motion and requires only a few frames. In RetinaDiff, registration based on phase correlation is first applied to stabilize the raw speckle sequence before contrast computation, reducing interframe misalignment so that fluctuations at each pixel primarily reflect true flow dynamics. This step provides a physics prior corrected for motion and a high quality multiframe tLSCI reference. Next, guided by the physics prior, a conditional diffusion model performs inverse reconstruction by jointly conditioning on the registered speckle sequence and the corrected prior. Experiments on data acquired with a retinal LSCI system developed in house show improved structural continuity and statistical stability compared with direct reconstruction from few frames and representative baselines. The framework also remains effective in a small number of extremely challenging cases, where both the direct 5-frame input and the conventional multiframe reconstruction are severely degraded. Overall, this work provides a practical and physically grounded route for reliable retinal tLSCI reconstruction from extremely limited frames. The source code and model weights will be publicly available at https://github.com/QianChen113/RetinaDiff.

JiaKui Hu, Yuxiao Yang, Jialun Liu et al.

Generating multi-view images from human instructions is crucial for 3D content creation. The primary challenges involve maintaining consistency across multiple views and effectively synthesizing shapes and textures under diverse conditions. In this paper, we propose the Multi-View Auto-Regressive (\textbf{MV-AR}) method, which leverages an auto-regressive model to progressively generate consistent multi-view images from arbitrary prompts. Firstly, the next-token-prediction capability of the AR model significantly enhances its effectiveness in facilitating progressive multi-view synthesis. When generating widely-separated views, MV-AR can utilize all its preceding views to extract effective reference information. Subsequently, we propose a unified model that accommodates various prompts via architecture designing and training strategies. To address multiple conditions, we introduce condition injection modules for text, camera pose, image, and shape. To manage multi-modal conditions simultaneously, a progressive training strategy is employed. This strategy initially adopts the text-to-multi-view (t2mv) model as a baseline to enhance the development of a comprehensive X-to-multi-view (X2mv) model through the randomly dropping and combining conditions. Finally, to alleviate the overfitting problem caused by limited high-quality data, we propose the ``Shuffle View" data augmentation technique, thus significantly expanding the training data by several magnitudes. Experiments demonstrate the performance and versatility of our MV-AR, which consistently generates consistent multi-view images across a range of conditions and performs on par with leading diffusion-based multi-view image generation models. The code and models are released at https://github.com/MILab-PKU/MVAR.

Hangzhou He, Jiachen Tang, Lei Zhu et al. · pku

Deep learning-based medical image classification techniques are rapidly advancing in medical image analysis, making it crucial to develop accurate and trustworthy models that can be efficiently deployed across diverse clinical scenarios. Concept Bottleneck Models (CBMs), which first predict a set of explainable concepts from images and then perform classification based on these concepts, are increasingly being adopted for explainable medical image classification. However, the inherent explainability of CBMs introduces new challenges when deploying trained models to new environments. Variations in imaging protocols and staining methods may induce concept-level shifts, such as alterations in color distribution and scale. Furthermore, since CBM training requires explicit concept annotations, fine-tuning models solely with image-level labels could compromise concept prediction accuracy and faithfulness - a critical limitation given the high cost of acquiring expert-annotated concept labels in medical domains. To address these challenges, we propose a training-free confusion concept identification strategy. By leveraging minimal new data (e.g., 4 images per class) with only image-level labels, our approach enhances out-of-domain performance without sacrificing source domain accuracy through two key operations: masking misactivated confounding concepts and amplifying under-activated discriminative concepts. The efficacy of our method is validated on both skin and white blood cell images. Our code is available at: https://github.com/riverback/TF-TTI-XMed.

Shuang Zeng, Lei Zhu, Xinliang Zhang et al. · pku

Medical image segmentation is a critical yet challenging task, primarily due to the difficulty of obtaining extensive datasets of high-quality, expert-annotated images. Contrastive learning presents a potential but still problematic solution to this issue. Because most existing methods focus on extracting instance-level or pixel-to-pixel representation, which ignores the characteristics between intra-image similar pixel groups. Moreover, when considering contrastive pairs generation, most SOTA methods mainly rely on manually setting thresholds, which requires a large number of gradient experiments and lacks efficiency and generalization. To address these issues, we propose a novel contrastive learning approach named SuperCL for medical image segmentation pre-training. Specifically, our SuperCL exploits the structural prior and pixel correlation of images by introducing two novel contrastive pairs generation strategies: Intra-image Local Contrastive Pairs (ILCP) Generation and Inter-image Global Contrastive Pairs (IGCP) Generation. Considering superpixel cluster aligns well with the concept of contrastive pairs generation, we utilize the superpixel map to generate pseudo masks for both ILCP and IGCP to guide supervised contrastive learning. Moreover, we also propose two modules named Average SuperPixel Feature Map Generation (ASP) and Connected Components Label Generation (CCL) to better exploit the prior structural information for IGCP. Finally, experiments on 8 medical image datasets indicate our SuperCL outperforms existing 12 methods. i.e. Our SuperCL achieves a superior performance with more precise predictions from visualization figures and 3.15%, 5.44%, 7.89% DSC higher than the previous best results on MMWHS, CHAOS, Spleen with 10% annotations. Our code will be released after acceptance.

Xinliang Zhang, Lei Zhu, Hangzhou He et al.

Multimodal Large Language Models (MLLMs) have demonstrated substantial value in unified text-image understanding and reasoning, primarily by converting images into sequences of patch-level tokens that align with their architectural paradigm. However, patch-level tokenization leads to a quadratic growth in image tokens, burdening MLLMs' understanding and reasoning with enormous computation and memory. Additionally, the traditional patch-wise scanning tokenization workflow misaligns with the human vision cognition system, further leading to hallucination and computational redundancy. To address this issue, we propose an object-level token merging strategy for Adaptive Token compression, revealing the consistency with human vision system. The experiments are conducted on multiple comprehensive benchmarks, which show that our approach averagely, utilizes only 10% tokens while achieving almost 96% of the vanilla model's performance. More extensive experimental results in comparison with relevant works demonstrate the superiority of our method in balancing compression ratio and performance. Our code will be available.

JiaKui Hu, Zhengjian Yao, Lujia Jin et al.

This study introduces a Masked Degradation Classification Pre-Training method (MaskDCPT), designed to facilitate the classification of degradation types in input images, leading to comprehensive image restoration pre-training. Unlike conventional pre-training methods, MaskDCPT uses the degradation type of the image as an extremely weak supervision, while simultaneously leveraging the image reconstruction to enhance performance and robustness. MaskDCPT includes an encoder and two decoders: the encoder extracts features from the masked low-quality input image. The classification decoder uses these features to identify the degradation type, whereas the reconstruction decoder aims to reconstruct a corresponding high-quality image. This design allows the pre-training to benefit from both masked image modeling and contrastive learning, resulting in a generalized representation suited for restoration tasks. Benefit from the straightforward yet potent MaskDCPT, the pre-trained encoder can be used to address universal image restoration and achieve outstanding performance. Implementing MaskDCPT significantly improves performance for both convolution neural networks (CNNs) and Transformers, with a minimum increase in PSNR of 3.77 dB in the 5D all-in-one restoration task and a 34.8% reduction in PIQE compared to baseline in real-world degradation scenarios. It also emergences strong generalization to previously unseen degradation types and levels. In addition, we curate and release the UIR-2.5M dataset, which includes 2.5 million paired restoration samples across 19 degradation types and over 200 degradation levels, incorporating both synthetic and real-world data. The dataset, source code, and models are available at https://github.com/MILab-PKU/MaskDCPT.

Shuang Zeng, Chee Hong Lee, Kaiwen Li et al. · pku

Retinal vessel segmentation plays a vital role in analyzing fundus images for the diagnosis of systemic and ocular diseases. Building on this, classifying segmented vessels into arteries and veins (A/V) further enables the extraction of clinically relevant features such as vessel width, diameter and tortuosity, which are essential for detecting conditions like diabetic and hypertensive retinopathy. However, manual segmentation and classification are time-consuming, costly and inconsistent. With the advancement of Convolutional Neural Networks, several automated methods have been proposed to address this challenge, but there are still some issues. For example, the existing methods all treat artery, vein and overall vessel segmentation as three separate binary tasks, neglecting the intrinsic coupling relationships between these anatomical structures. Considering artery and vein structures are subsets of the overall retinal vessel map and should naturally exhibit prediction consistency with it, we design a novel loss named Channel-Coupled Vessel Consistency Loss to enforce the coherence and consistency between vessel, artery and vein predictions, avoiding biasing the network toward three simple binary segmentation tasks. Moreover, we also introduce a regularization term named intra-image pixel-level contrastive loss to extract more discriminative feature-level fine-grained representations for accurate retinal A/V classification. SOTA results have been achieved across three public A/V classification datasets including RITE, LES-AV and HRF. Our code will be available upon acceptance.

Lei Zhu, Fangyun Wei, Yanye Lu et al.

In the realm of image quantization exemplified by VQGAN, the process encodes images into discrete tokens drawn from a codebook with a predefined size. Recent advancements, particularly with LLAMA 3, reveal that enlarging the codebook significantly enhances model performance. However, VQGAN and its derivatives, such as VQGAN-FC (Factorized Codes) and VQGAN-EMA, continue to grapple with challenges related to expanding the codebook size and enhancing codebook utilization. For instance, VQGAN-FC is restricted to learning a codebook with a maximum size of 16,384, maintaining a typically low utilization rate of less than 12% on ImageNet. In this work, we propose a novel image quantization model named VQGAN-LC (Large Codebook), which extends the codebook size to 100,000, achieving an utilization rate exceeding 99%. Unlike previous methods that optimize each codebook entry, our approach begins with a codebook initialized with 100,000 features extracted by a pre-trained vision encoder. Optimization then focuses on training a projector that aligns the entire codebook with the feature distributions of the encoder in VQGAN-LC. We demonstrate the superior performance of our model over its counterparts across a variety of tasks, including image reconstruction, image classification, auto-regressive image generation using GPT, and image creation with diffusion- and flow-based generative models. Code and models are available at https://github.com/zh460045050/VQGAN-LC.

Shuang Zeng, Chee Hong Lee, Micky C Nnamdi et al. · pku

Retinal vessel segmentation is a vital early detection method for several severe ocular diseases. Despite significant progress in retinal vessel segmentation with the advancement of Neural Networks, there are still challenges to overcome. Specifically, retinal vessel segmentation aims to predict the class label for every pixel within a fundus image, with a primary focus on intra-image discrimination, making it vital for models to extract more discriminative features. Nevertheless, existing methods primarily focus on minimizing the difference between the output from the decoder and the label, but ignore fully using feature-level fine-grained representations from the encoder. To address these issues, we propose a novel Attention U-shaped Kolmogorov-Arnold Network named AttUKAN along with a novel Label-guided Pixel-wise Contrastive Loss for retinal vessel segmentation. Specifically, we implement Attention Gates into Kolmogorov-Arnold Networks to enhance model sensitivity by suppressing irrelevant feature activations and model interpretability by non-linear modeling of KAN blocks. Additionally, we also design a novel Label-guided Pixel-wise Contrastive Loss to supervise our proposed AttUKAN to extract more discriminative features by distinguishing between foreground vessel-pixel pairs and background pairs. Experiments are conducted across four public datasets including DRIVE, STARE, CHASE_DB1, HRF and our private dataset. AttUKAN achieves F1 scores of 82.50%, 81.14%, 81.34%, 80.21% and 80.09%, along with MIoU scores of 70.24%, 68.64%, 68.59%, 67.21% and 66.94% in the above datasets, which are the highest compared to 11 networks for retinal vessel segmentation. Quantitative and qualitative results show that our AttUKAN achieves state-of-the-art performance and outperforms existing retinal vessel segmentation methods. Our code will be available at https://github.com/stevezs315/AttUKAN.

Zhenxin Lei, Man Yao, Jiakui Hu et al.

Spiking Neural Networks (SNNs) have a low-power advantage but perform poorly in image segmentation tasks. The reason is that directly converting neural networks with complex architectural designs for segmentation tasks into spiking versions leads to performance degradation and non-convergence. To address this challenge, we first identify the modules in the architecture design that lead to the severe reduction in spike firing, make targeted improvements, and propose Spike2Former architecture. Second, we propose normalized integer spiking neurons to solve the training stability problem of SNNs with complex architectures. We set a new state-of-the-art for SNNs in various semantic segmentation datasets, with a significant improvement of +12.7% mIoU and 5.0 efficiency on ADE20K, +14.3% mIoU and 5.2 efficiency on VOC2012, and +9.1% mIoU and 6.6 efficiency on CityScapes.

Hangzhou He, Lei Zhu, Xinliang Zhang et al. · pku

Concept Bottleneck Models (CBMs) offer inherent interpretability by initially translating images into human-comprehensible concepts, followed by a linear combination of these concepts for classification. However, the annotation of concepts for visual recognition tasks requires extensive expert knowledge and labor, constraining the broad adoption of CBMs. Recent approaches have leveraged the knowledge of large language models to construct concept bottlenecks, with multimodal models like CLIP subsequently mapping image features into the concept feature space for classification. Despite this, the concepts produced by language models can be verbose and may introduce non-visual attributes, which hurts accuracy and interpretability. In this study, we investigate to avoid these issues by constructing CBMs directly from multimodal models. To this end, we adopt common words as base concept vocabulary and leverage auxiliary unlabeled images to construct a Vision-to-Concept (V2C) tokenizer that can explicitly quantize images into their most relevant visual concepts, thus creating a vision-oriented concept bottleneck tightly coupled with the multimodal model. This leads to our V2C-CBM which is training efficient and interpretable with high accuracy. Our V2C-CBM has matched or outperformed LLM-supervised CBMs on various visual classification benchmarks, validating the efficacy of our approach.

Xinliang Zhang, Lei Zhu, Shuang Zeng et al. · pku

Scribble-based weakly supervised semantic segmentation leverages only a few annotated pixels as labels to train a segmentation model, presenting significant potential for reducing the human labor involved in the annotation process. This approach faces two primary challenges: first, the sparsity of scribble annotations can lead to inconsistent predictions due to limited supervision; second, the variability in scribble annotations, reflecting differing human annotator preferences, can prevent the model from consistently capturing the discriminative regions of objects, potentially leading to unstable predictions. To address these issues, we propose a holistic framework, the class-driven scribble promotion network, for robust scribble-supervised semantic segmentation. This framework not only utilizes the provided scribble annotations but also leverages their associated class labels to generate reliable pseudo-labels. Within the network, we introduce a localization rectification module to mitigate noisy labels and a distance perception module to identify reliable regions surrounding scribble annotations and pseudo-labels. In addition, we introduce new large-scale benchmarks, ScribbleCOCO and ScribbleCityscapes, accompanied by a scribble simulation algorithm that enables evaluation across varying scribble styles. Our method demonstrates competitive performance in both accuracy and robustness, underscoring its superiority over existing approaches. The datasets and the codes will be made publicly available.

Zhengjian Yao, Yongzhi Li, Xinyuan Gao et al.

We present "Narrative Weaver", a novel framework that addresses a fundamental challenge in generative AI: achieving multi-modal controllable, long-range, and consistent visual content generation. While existing models excel at generating high-fidelity short-form visual content, they struggle to maintain narrative coherence and visual consistency across extended sequences - a critical limitation for real-world applications such as filmmaking and e-commerce advertising. Narrative Weaver introduces the first holistic solution that seamlessly integrates three essential capabilities: fine-grained control, automatic narrative planning, and long-range coherence. Our architecture combines a Multimodal Large Language Model (MLLM) for high-level narrative planning with a novel fine-grained control module featuring a dynamic Memory Bank that prevents visual drift. To enable practical deployment, we develop a progressive, multi-stage training strategy that efficiently leverages existing pre-trained models, achieving state-of-the-art performance even with limited training data. Recognizing the absence of suitable evaluation benchmarks, we construct and release the E-commerce Advertising Video Storyboard Dataset (EAVSD) - the first comprehensive dataset for this task, containing over 330K high-quality images with rich narrative annotations. Through extensive experiments across three distinct scenarios (controllable multi-scene generation, autonomous storytelling, and e-commerce advertising), we demonstrate our method's superiority while opening new possibilities for AI-driven content creation.

JiaKui Hu, Zhengjian Yao, Lujia Jin et al. · pku

Translation equivariance is a fundamental inductive bias in image restoration, ensuring that translated inputs produce translated outputs. Attention mechanisms in modern restoration transformers undermine this property, adversely impacting both training convergence and generalization. To alleviate this issue, we propose two key strategies for incorporating translation equivariance: slide indexing and component stacking. Slide indexing maintains operator responses at fixed positions, with sliding window attention being a notable example, while component stacking enables the arrangement of translation-equivariant operators in parallel or sequentially, thereby building complex architectures while preserving translation equivariance. However, these strategies still create a dilemma in model design between the high computational cost of self-attention and the fixed receptive field associated with sliding window attention. To address this, we develop an adaptive sliding indexing mechanism to efficiently select key-value pairs for each query, which are then concatenated in parallel with globally aggregated key-value pairs. The designed network, called the Translation Equivariance Adaptive Transformer (TEAFormer), is assessed across a variety of image restoration tasks. The results highlight its superiority in terms of effectiveness, training convergence, and generalization.

Hangzhou He, Lei Zhu, Kaiwen Li et al. · pku

Concept Bottleneck Models (CBMs) provide inherent interpretability by first predicting a set of human-understandable concepts and then mapping them to labels through a simple classifier. While users can intervene in the concept space to improve predictions, traditional CBMs typically employ a fixed linear classifier over concept scores, which restricts interventions to manual value adjustments and prevents the incorporation of new concepts or domain knowledge at test time. These limitations are particularly severe in unsupervised CBMs, where concept activations are often noisy and densely activated, making user interventions ineffective. We introduce Chat-CBM, which replaces score-based classifiers with a language-based classifier that reasons directly over concept semantics. By grounding prediction in the semantic space of concepts, Chat-CBM preserves the interpretability of CBMs while enabling richer and more intuitive interventions, such as concept correction, addition or removal of concepts, incorporation of external knowledge, and high-level reasoning guidance. Leveraging the language understanding and few-shot capabilities of frozen large language models, Chat-CBM extends the intervention interface of CBMs beyond numerical editing and remains effective even in unsupervised settings. Experiments on nine datasets demonstrate that Chat-CBM achieves higher predictive performance and substantially improves user interactivity while maintaining the concept-based interpretability of CBMs.

Ourui Fu, Hangzhou He, Xinliang Zhang et al. · pku

The annotation bottleneck in semantic segmentation has driven significant interest in few-shot segmentation, which aims to develop segmentation models capable of generalizing rapidly to novel classes using minimal exemplars. Conventional training paradigms typically generate query prior maps by extracting masked-area features from support images, followed by making predictions guided by these prior maps. However, current approaches remain constrained by two critical limitations stemming from inter- and intra-image discrepancies, both of which significantly degrade segmentation performance: 1) The semantic gap between support and query images results in mismatched features and inaccurate prior maps; 2) Visually similar yet semantically distinct regions within support or query images lead to false negative or false positive predictions. We propose a novel FSS method called \textbf{I$^2$R}: 1) Using category-specific high level representations which aggregate global semantic cues from support and query images, enabling more precise inter-image region localization and address the first limitation. 2) Directional masking strategy that suppresses inconsistent support-query pixel pairs, which exhibit high feature similarity but conflicting mask, to mitigate the second issue. Experiments demonstrate that our method outperforms state-of-the-art approaches, achieving improvements of 1.9\% and 2.1\% in mIoU under the 1-shot setting on PASCAL-5$^i$ and COCO-20$^i$ benchmarks, respectively.

Qian Chen, Lei Zhu, Hangzhou He et al.

The primary goal of continual learning (CL) task in medical image segmentation field is to solve the "catastrophic forgetting" problem, where the model totally forgets previously learned features when it is extended to new categories (class-level) or tasks (task-level). Due to the privacy protection, the historical data labels are inaccessible. Prevalent continual learning methods primarily focus on generating pseudo-labels for old datasets to force the model to memorize the learned features. However, the incorrect pseudo-labels may corrupt the learned feature and lead to a new problem that the better the model is trained on the old task, the poorer the model performs on the new tasks. To avoid this problem, we propose a network by introducing the data-specific Mixture of Experts (MoE) structure to handle the new tasks or categories, ensuring that the network parameters of previous tasks are unaffected or only minimally impacted. To further overcome the tremendous memory costs caused by introducing additional structures, we propose a Low-Rank strategy which significantly reduces memory cost. We validate our method on both class-level and task-level continual learning challenges. Extensive experiments on multiple datasets show our model outperforms all other methods.

Lei Zhu, Qi She, Qian Chen et al.

Weakly supervised object localization (WSOL) relaxes the requirement of dense annotations for object localization by using image-level classification masks to supervise its learning process. However, current WSOL methods suffer from excessive activation of background locations and need post-processing to obtain the localization mask. This paper attributes these issues to the unawareness of background cues, and propose the background-aware classification activation map (B-CAM) to simultaneously learn localization scores of both object and background with only image-level labels. In our B-CAM, two image-level features, aggregated by pixel-level features of potential background and object locations, are used to purify the object feature from the object-related background and to represent the feature of the pure-background sample, respectively. Then based on these two features, both the object classifier and the background classifier are learned to determine the binary object localization mask. Our B-CAM can be trained in end-to-end manner based on a proposed stagger classification loss, which not only improves the objects localization but also suppresses the background activation. Experiments show that our B-CAM outperforms one-stage WSOL methods on the CUB-200, OpenImages and VOC2012 datasets.

Lei Zhu, Qi She, Duo Li et al.

The nonlocal-based blocks are designed for capturing long-range spatial-temporal dependencies in computer vision tasks. Although having shown excellent performance, they still lack the mechanism to encode the rich, structured information among elements in an image or video. In this paper, to theoretically analyze the property of these nonlocal-based blocks, we provide a new perspective to interpret them, where we view them as a set of graph filters generated on a fully-connected graph. Specifically, when choosing the Chebyshev graph filter, a unified formulation can be derived for explaining and analyzing the existing nonlocal-based blocks (e.g., nonlocal block, nonlocal stage, double attention block). Furthermore, by concerning the property of spectral, we propose an efficient and robust spectral nonlocal block, which can be more robust and flexible to catch long-range dependencies when inserted into deep neural networks than the existing nonlocal blocks. Experimental results demonstrate the clear-cut improvements and practical applicabilities of our method on image classification, action recognition, semantic segmentation, and person re-identification tasks.

Mengdi Gao, Ximeng Feng, Mufeng Geng et al.

Purpose: Deep neural networks (DNNs) have been widely applied in medical image classification, benefiting from its powerful mapping capability among medical images. However, these existing deep learning-based methods depend on an enormous amount of carefully labeled images. Meanwhile, noise is inevitably introduced in the labeling process, degrading the performance of models. Hence, it's significant to devise robust training strategies to mitigate label noise in the medical image classification tasks. Methods: In this work, we propose a novel Bayesian statistics guided label refurbishment mechanism (BLRM) for DNNs to prevent overfitting noisy images. BLRM utilizes maximum a posteriori probability (MAP) in the Bayesian statistics and the exponentially time-weighted technique to selectively correct the labels of noisy images. The training images are purified gradually with the training epochs when BLRM is activated, further improving classification performance. Results: Comprehensive experiments on both synthetic noisy images (public OCT & Messidor datasets) and real-world noisy images (ANIMAL-10N) demonstrate that BLRM refurbishes the noisy labels selectively, curbing the adverse effects of noisy data. Also, the anti-noise BLRM integrated with DNNs are effective at different noise ratio and are independent of backbone DNN architectures. In addition, BLRM is superior to state-of-the-art comparative methods of anti-noise. Conclusions: These investigations indicate that the proposed BLRM is well capable of mitigating label noise in medical image classification tasks.

Lei Zhu, Qi She, Bin Zhang et al.

Superpixel is generated by automatically clustering pixels in an image into hundreds of compact partitions, which is widely used to perceive the object contours for its excellent contour adherence. Although some works use the Convolution Neural Network (CNN) to generate high-quality superpixel, we challenge the design principles of these networks, specifically for their dependence on manual labels and excess computation resources, which limits their flexibility compared with the traditional unsupervised segmentation methods. We target at redefining the CNN-based superpixel segmentation as a lifelong clustering task and propose an unsupervised CNN-based method called LNS-Net. The LNS-Net can learn superpixel in a non-iterative and lifelong manner without any manual labels. Specifically, a lightweight feature embedder is proposed for LNS-Net to efficiently generate the cluster-friendly features. With those features, seed nodes can be automatically assigned to cluster pixels in a non-iterative way. Additionally, our LNS-Net can adapt the sequentially lifelong learning by rescaling the gradient of weight based on both channel and spatial context to avoid overfitting. Experiments show that the proposed LNS-Net achieves significantly better performance on three benchmarks with nearly ten times lower complexity compared with other state-of-the-art methods.