论文检索

Ji Ma, Wei Suo, Peng Wang, Yanning Zhang

Vision-Language Instruction Tuning (VLIT) is a critical training phase for Large Vision-Language Models (LVLMs). With the improving capabilities of open-source LVLMs, researchers have increasingly turned to generate VLIT data by using open-source LVLMs and achieved significant progress. However, such data generation approaches are bottlenecked by the following challenges: 1) Since multi-modal models tend to be influenced by prior language knowledge, directly using LVLMs to generate VLIT data would inevitably lead to low content relevance between generated data and images. 2) To improve the ability of the models to generate VLIT data, previous methods have incorporated an additional training phase to boost the generative capacity. This process hurts the generalization of the models to unseen inputs (i.e., “exposure bias” problem). In this paper, we propose a new Content Correlated VLIT data generation via Contrastive Learning (C3L). Specifically, we design a new content relevance module which enhances the content relevance between VLIT data and images by computing Image Instruction Correspondence Scores S(I2C). Moreover, a contrastive learning module is introduced to further boost the VLIT data generation capability of the LVLMs. A large number of automatic measures on four benchmarks show the effectiveness of our method.

Yingnan Liu, Yingtian Zou, Rui Qiao, Fusheng Liu, Mong Li Lee, Wynne Hsu

Domain generalization aims to develop models that are robust to distribution shifts. Existing methods focus on learning invariance across domains to enhance model robustness, and data augmentation has been widely used to learn invariant predictors, with most methods performing augmentation in the input space. However, augmentation in the input space has limited diversity whereas in the feature space is more versatile and has shown promising results. Nonetheless, feature semantics is seldom considered and existing feature augmentation methods suffer from a limited variety of augmented features. We decompose features into class-generic, class-specific, domain-generic, and domain-specific components. We propose a cross-domain feature augmentation method named XDomainMix that enables us to increase sample diversity while emphasizing the learning of invariant representations to achieve domain generalization. Experiments on widely used benchmark datasets demonstrate that our proposed method is able to achieve state-of-the-art performance. Quantitative analysis indicates that our feature augmentation approach facilitates the learning of effective models that are invariant across different domains.

Yaohua Liu, Jiaxin Gao, Xuan Liu, Xianghao Jiao, Xin Fan, Risheng Liu

Transfer attacks generate significant interest for real-world black-box applications by crafting transferable adversarial examples through surrogate models. Whereas, existing works essentially directly optimize the single-level objective w.r.t. the surrogate model, which always leads to poor interpretability of attack mechanism and limited generalization performance over unknown victim models. In this work, we propose the BilEvel Transfer AttacK (BETAK) framework by establishing an initialization derived bilevel optimization paradigm, which explicitly reformulates the nested constraint relationship between the Upper-Level (UL) pseudo-victim attacker and the Lower-Level (LL) surrogate attacker. Algorithmically, we introduce the Hyper Gradient Response (HGR) estimation as an effective feedback for the transferability over pseudo-victim attackers, and propose the Dynamic Sequence Truncation (DST) technique to dynamically adjust the back-propagation path for HGR and reduce computational overhead simultaneously. Meanwhile, we conduct detailed algorithmic analysis and provide convergence guarantee to support non-convexity of the LL surrogate attacker. Extensive evaluations demonstrate substantial improvement of BETAK (e.g., 53.41% increase of attack success rates against IncRes-v2_ens victim) against different victims and defense methods in targeted and untargeted attack scenarios.

Yanghong Liu, Xingping Dong, Yutian Lin, Mang Ye

Recent years have witnessed the success of generative adversarial networks and diffusion models in multi-model trajectory prediction. However, prevailing algorithms only explicitly consider human interaction, but ignore the modeling of human intention, yielding that the generated results deviate largely from real trajectories in some complex scenes. In this paper, we analyze the conditions of multi-modal trajectory prediction from two objective perspectives and propose a novel end-to-end framework based on the diffusion model to predict more precise and socially-acceptable trajectories for humans. Firstly, a spatial-temporal aggregation module is built to extract the extrinsic interaction features for capturing socially-acceptable behaviors. Secondly, we explicitly construct the intrinsic intention module to obtain intention features for precise prediction. Finally, we estimate a noise trajectory distribution with these two features as the initiation of diffusion model and leverage denoising process to obtain the final trajectories. Furthermore, to reduce the noise of the initiative trajectory estimation, we present a novel sample consistency loss to constrain multiple predictions. Extensive experiments demonstrate that our method outperforms the state-of-the-art methods on ETH-UCY and SDD benchmarks, specifically achieving 19.0%/24.2% ADE/FDE improvement on ETH-UCY.

Shuai Liu, Boyang Li, Zhiyu Fang, Kai Huang

Recently, significant progress has been made in the research of 3D object detection. However, most prior studies have focused on the utilization of center-based or anchor-based label assignment schemes. Alternative label assignment strategies remain unexplored in 3D object detection. We find that the center-based label assignment often fails to generate sufficient positive samples for training, while the anchor-based label assignment tends to encounter an imbalanced issue when handling objects with different scales. To solve these issues, we introduce a dynamic cross label assignment (DCLA) scheme, which dynamically assigns positive samples for each object from a cross-shaped region, thus providing sufficient and balanced positive samples for training. Furthermore, to address the challenge of accurately regressing objects with varying scales, we put forth a rotation-weighted Intersection over Union (RWIoU) metric to replace the widely used L1 metric in regression loss. Extensive experiments demonstrate the generality and effectiveness of our DCLA and RWIoU-based regression loss. The Code is available at https://github.com/Say2L/DCDet.git.

Jinyuan Liu, Guanyao Wu, Zhu Liu, Long Ma, Risheng Liu, Xin Fan

Multi-modality image fusion aims to integrate images from multiple sensors, producing an image that is visually appealing and offers more comprehensive information than any single one. To ensure high visual quality and facilitate accurate subsequent perception tasks, previous methods have often cascaded networks using weighted loss functions. However, such simplistic strategies struggle to truly achieve the "Best of Both Worlds", and the adjustment of numerous hand-crafted parameters becomes burdensome. To address these challenges, this paper introduces a Compact, Automatic and Flexible framework, dubbed CAF, designed for infrared and visible image fusion, along with subsequent tasks. Concretely, we recast the combined problem of fusion and perception into a single objective, allowing mutual optimization of information from both tasks. Then we also utilize the perception task to inform the design of fusion loss functions, facilitating the automatic identification of optimal fusion objectives tailored to the task. Furthermore, CAF can support seamless integration with existing approaches easily, offering flexibility in adapting to various tasks and network structures. Extensive experiments demonstrate the superiority of CAF, which not only produces visually admirable fused results but also realizes 1.7 higher detection mAP@.5 and 2.0 higher segmentation mIoU than the state-of-the-art methods. The code is available at https://github.com/RollingPlain/CAF_IVIF.

Gongye Liu, Haoze Sun, Jiayi Li, Fei Yin, Yujiu Yang

Diffusion models have recently demonstrated an impressive ability to address inverse problems in an unsupervised manner. While existing methods primarily focus on modifying the posterior sampling process, the potential of the forward process remains largely unexplored. In this work, we propose Shortcut Sampling for Diffusion(SSD), a novel approach for solving inverse problems in a zero-shot manner. Instead of initiating from random noise, the core concept of SSD is to find a specific transitional state that bridges the measurement image y and the restored image x. By utilizing the shortcut path of "input - transitional state - output", SSD can achieve precise restoration with fewer steps. To derive the transitional state during the forward process, we introduce Distortion Adaptive Inversion. Moreover, we apply back projection as additional consistency constraints during the generation process. Experimentally, we demonstrate SSD's effectiveness on multiple representative IR tasks. Our method achieves competitive results with only 30 NFEs compared to state-of-the-art zero-shot methods(100 NFEs) and outperforms them with 100 NFEs in certain tasks. Code is available at https://github.com/GongyeLiu/SSD.

Chuni Liu, Boyuan Ma, Xiaojuan Ban, Yujie Xie, Hao Wang, Weihua Xue, Jingchao Ma, Ke Xu

Topological consistency plays a crucial role in the task of boundary segmentation for reticular images, such as cell membrane segmentation in neuron electron microscopic images, grain boundary segmentation in material microscopic images and road segmentation in aerial images. In these fields, topological changes in segmentation results have a serious impact on the downstream tasks, which can even exceed the misalignment of the boundary itself. To enhance the topology accuracy in segmentation results, we propose the Skea-Topo Aware loss, which is a novel loss function that takes into account the shape of each object and topological significance of the pixels. It consists of two components. First, a skeleton-aware weighted loss improves the segmentation accuracy by better modeling the object geometry with skeletons. Second, a boundary rectified term effectively identifies and emphasizes topological critical pixels in the prediction errors using both foreground and background skeletons in the ground truth and predictions. Experiments prove that our method improves topological consistency by up to 7 points in VI compared to 13 state-of-art methods, based on objective and subjective assessments across three different boundary segmentation datasets. The code is available at https://github.com/clovermini/Skea_topo.

Zeyu Ling, Bo Han, Yongkang Wong, Han Lin, Mohan Kankanhalli, Weidong Geng

Conditional human motion synthesis (HMS) aims to generate human motion sequences that conform to specific conditions. Text and audio represent the two predominant modalities employed as HMS control conditions. While existing research has primarily focused on single conditions, the multi-condition human motion synthesis remains underexplored. In this study, we propose a multi-condition HMS framework, termed MCM, based on a dual-branch structure composed of a main branch and a control branch. This framework effectively extends the applicability of the diffusion model, which is initially predicated solely on textual conditions, to auditory conditions. This extension encompasses both music-to-dance and co-speech HMS while preserving the intrinsic quality of motion and the capabilities for semantic association inherent in the original model. Furthermore, we propose the implementation of a Transformer-based diffusion model, designated as MWNet, as the main branch. This model adeptly apprehends the spatial intricacies and inter-joint correlations inherent in motion sequences, facilitated by the integration of multi-wise self-attention modules. Extensive experiments show that our method achieves competitive results in single-condition and multi-condition HMS tasks.

Quanmin Liang, Zhilin Huang, Xiawu Zheng, Feidiao Yang, Jun Peng, Kai Huang, Yonghong Tian

Current Event Stream Super-Resolution (ESR) methods overlook the redundant and complementary information present in positive and negative events within the event stream, employing a direct mixing approach for super-resolution, which may lead to detail loss and inefficiency. To address these issues, we propose an efficient Recursive Multi-Branch Information Fusion Network (RMFNet) that separates positive and negative events for complementary information extraction, followed by mutual supplementation and refinement. Particularly, we introduce Feature Fusion Modules (FFM) and Feature Exchange Modules (FEM). FFM is designed for the fusion of contextual information within neighboring event streams, leveraging the coupling relationship between positive and negative events to alleviate the misleading of noises in the respective branches. FEM efficiently promotes the fusion and exchange of information between positive and negative branches, enabling superior local information enhancement and global information complementation. Experimental results demonstrate that our approach achieves over 17% and 31% improvement on synthetic and real datasets, accompanied by a 2.3x acceleration. Furthermore, we evaluate our method on two downstream event-driven applications, i.e., object recognition and video reconstruction, achieving remarkable results that outperform existing methods. Our code and Supplementary Material are available at https://github.com/Lqm26/RMFNet.

Jiafeng Liang, Shixin Jiang, Zekun Wang, Haojie Pan, Zerui Chen, Zheng Chu, Ming Liu, Ruiji Fu, Zhongyuan Wang, Bing Qin

There are substantial instructional videos on the Internet, which provide us tutorials for completing various tasks. Existing instructional video datasets only focus on specific steps at the video level, lacking experiential guidelines at the task level, which can lead to beginners struggling to learn new tasks due to the lack of relevant experience. Moreover, the specific steps without guidelines are trivial and unsystematic, making it difficult to provide a clear tutorial. To address these problems, we present the Guide (Guideline-Guided) dataset, which contains 3.5K videos of 560 instructional tasks in 8 domains related to our daily life. Specifically, we annotate each instructional task with a guideline, representing a common pattern shared by all task-related videos. On this basis, we annotate systematic specific steps, including their associated guideline steps, specific step descriptions and timestamps. Our proposed benchmark consists of three sub-tasks to evaluate comprehension ability of models: (1) Step Captioning: models have to generate captions for specific steps from videos. (2) Guideline Summarization: models have to mine the common pattern in task-related videos and summarize a guideline from them. (3) Guideline-Guided Captioning: models have to generate captions for specific steps under the guide of guideline. We evaluate plenty of foundation models with Guide and perform in-depth analysis. Given the diversity and practicality of Guide, we believe that it can be used as a better benchmark for instructional video comprehension.

Guoyan Liang, Qin Zhou, Jingyuan Chen, Zhe Wang, Chang Yao

Medical Image Segmentation (MIS) plays a crucial role in medical therapy planning and robot navigation. Prototype learning methods in MIS focus on generating segmentation masks through pixel-to-prototype comparison. However, current approaches often overlook sample diversity by using a fixed prototype per semantic class and neglect intra-class variation within each input. In this paper, we propose to generate instance-adaptive prototypes for MIS, which integrates a common prototype proposal (CPP) capturing common visual patterns and an instance-specific prototype proposal (IPP) tailored to each input. To further account for the intra-class variation, we propose to guide the IPP generation by re-weighting the intermediate feature map according to their confidence scores. These confidence scores are hierarchically generated using a transformer decoder. Additionally we introduce a novel self-supervised filtering strategy to prioritize the foreground pixels during the training of the transformer decoder. Extensive experiments demonstrate favorable performance of our method.

Zongyi Li, Jianbo Li, Yuxuan Shi, Hefei Ling, Jiazhong Chen, Runsheng Wang, Shijuan Huang

Text-based Person Search aims to retrieve a specified person using a given text query. Current methods predominantly rely on paired labeled image-text data to train the cross-modality retrieval model, necessitating laborious and time-consuming labeling. In response to this challenge, we present the Cross-modal Generation and Alignment via Attribute-guided Prompt framework (GAAP) for fully unsupervised text-based person search, utilizing only unlabeled images. Our proposed GAAP framework consists of two key parts: Attribute-guided Prompt Caption Generation and Attribute-guided Cross-modal Alignment module. The Attribute-guided Prompt Caption Generation module generates pseudo text labels by feeding the attribute prompts into a large-scale pre-trained vision-language model. These synthetic texts are then meticulously selected through a sample selection, ensuring the reliability for subsequent fine-tuning. The Attribute-guided Cross-modal Alignment module encompasses three sub-modules for feature alignment across modalities. Firstly, Cross-Modal Center Alignment (CMCA) aligns the samples with different modality centroids. Subsequently, to address ambiguity arising from local attribute similarities, an Attribute-guided Image-Text Contrastive Learning module (AITC) is proposed to facilitate the alignment of relationships among different pairs by considering local attribute similarities. Lastly, the Attribute-guided Image-Text Matching (AITM) module is introduced to mitigate noise in pseudo captions by using the image-attribute matching score to soften the hard matching labels. Empirical results showcase the effectiveness of our method across various text-based person search datasets under the fully unsupervised setting.

Xuan Li, Xiangqian Wu

Diabetic retinopathy lesion segmentation (DRLS) faces a challenge of significant variation in the size of different lesions. An effective method to address this challenge is to fuse multi-scale features. To boost the performance of this kind of method, most existing DRLS methods work on devising sophisticated multi-scale feature fusion modules. Differently, we focus on improving the quality of the multi-scale features to enhance the fused multi-scale feature representation. To this end, we design a Wavelet-based Scale-specific Recurrent Feedback Network (WSRFNet), which refines multi-scale features using recurrent feedback mechanism. Specifically, to avoid information loss when introducing feedback to multi-scale features, we propose a wavelet-based feedback pyramid module (WFPM), which is based on a reversible downsampling operation, i.e., Haar wavelet transform. Unlike scale-agnostic feedback used in previous feedback methods, we develop a scale-specific refinement module (SRM), which utilizes scale-specific feedback to pointedly refine features of different scales. Experimental results on IDRiD and DDR datasets show that our approach outperforms state-of-the-art models. The code is available at https://github.com/xuanli01/WSRFNet.

Xin Li, Yuxin Feng, Fan Zhou, Yun Liang, Zhuo Su

Synthetic data-driven methods perform well on image rain removal task, but they still face many challenges in real rainfall scenarios due to the complexity and diversity of rainy patterns. In this paper, we propose a new generic paradigm for real image deraining from the perspective of synthesizing data covering more rainy patterns and constructing image rain removal networks with strong generalization performance. Firstly, instead of simply superimposing rain layers, we integrate various rainy patterns and design a phenomenal pipeline that incorporates multiple degradation types. Secondly, we construct a Patterns-aware Rain Removal Network (PRRN), which learns from both synthetic and real data simultaneously. In addition, to eliminate the inevitable distribution differences between synthetic and real data, we design a new Multi-representation Inter-domain Alignment Module (MIAM) in PRRN. By using multiple parallel submodules, MIAM achieves alignment of data domains in multiple feature subspaces. Based on several authoritative objective evaluation metrics, we successfully validate the effectiveness and robustness of the proposed method in real scenarios through extensive experiments carried out on five challenging real datasets.

Xianbang Li, Yilong Ren, Han Jiang, Haiyang Yu, Yanlei Cui, Liang Xu

Pedestrian trajectory prediction has emerged as a core component of human-robot interaction and autonomous driving. Fast and accurate prediction of surrounding pedestrians contributes to making decisions and improves safety and efficiency. However, pedestrians’ future trajectories will interact with their surrounding traffic participants. As the density of pedestrians increases, the complexity of such interactions also increases significantly, leading to an inevitable decrease in the accuracy of pedestrian trajectory prediction. To address this issue, we propose DenseKoopman, a plug-and-play framework for dense pedestrian trajectory prediction. Specifically, we introduce the Koopman operator theory to find an embedding space for a global linear approximation of a nonlinear pedestrian motion system. By encoding historical trajectories as linear state embeddings in the Koopman space, we transforms nonlinear trajectory data for pedestrians in dense scenes. This linearized representation greatly reduces the complexity of dense pedestrian trajectory prediction. Extensive experiments on pedestrian trajectory prediction benchmarks demonstrate the superiority of the proposed framework. We also conducted an analysis of the data transformation to explore how our DenseKoopman framework works with each validation method and uncovers motion patterns that may be hidden within the trajectory data. Code is available at https://github.com/lixianbang/DenseKoopman.

Pengteng Li, Ying He, F. Richard Yu, Pinhao Song, Xingchen Zhou, Guang Zhou

The autonomous driving community is highly interested in 3D occupancy prediction due to its outstanding geometric perception and object recognition capabilities. However, previous methods are limited to existing semantic conversion mechanisms for solving sparse ground truths problem, causing excessive computational demands and sub-optimal voxels representation. To tackle the above limitations, we propose OTOcc, a novel 3D occupancy prediction framework that models semantic conversion from 2D pixels to 3D voxels as Optimal Transport (OT) problem, offering accurate semantic mapping to adapt to sparse scenarios without attention or depth estimation. Specifically, the unit transportation cost between each demander (voxel) and supplier (pixel) pair is defined as the weighted occupancy prediction loss. Then, we utilize the Sinkhorn-Knopp Iteration to find the best mapping matrices with minimal transportation costs. To reduce the computational cost, we propose a block reading technique with multi-perspective feature representation, which also brings fine-grained scene understanding. Extensive experiments show that OTOcc not only has the competitive prediction performance but also has about more than 4.58% reduction in computational overhead compared to state-of-the-art methods.

Junjie Li, Yixin Zhang, Zilei Wang, Saihui Hou, Keyu Tu, Man Zhang

Contrastive learning has shown impressive success in enhancing feature discriminability for various visual tasks in a self-supervised manner, but the standard contrastive paradigm (features+l2 normalization) has limited benefits when applied in domain adaptation. We find that this is mainly because the class weights (weights of the final fully connected layer) are ignored in the domain adaptation optimization process, which makes it difficult for features to cluster around the corresponding class weights. To solve this problem, we propose the simple but powerful Probabilistic Contrastive Learning (PCL), which moves beyond the standard paradigm by removing l2 normalization and replacing the features with probabilities. PCL can guide the probability distribution towards a one-hot configuration, thus minimizing the discrepancy between features and class weights. We conduct extensive experiments to validate the effectiveness of PCL and observe consistent performance gains on five tasks, i.e., Unsupervised/Semi-Supervised Domain Adaptation (UDA/SSDA), Semi-Supervised Learning (SSL), UDA Detection and Semantic Segmentation. Notably, for UDA Semantic Segmentation on SYNTHIA, PCL surpasses the sophisticated CPSL-D by 2% in terms of mean IoU with a much lower training cost (PCL: 1*3090, 5 days v.s. CPSL-D: 4*V100, 11 days). Code is available at https://github.com/ljjcoder/Probabilistic-Contrastive-Learning.

Jinmin Li, Tao Dai, Jingyun Zhang, Kang Liu, Jun Wang, Shaoming Wang, Shu-Tao Xia, Rizen Guo

Recently developed generative methods, including invertible rescaling network (IRN) based and generative adversarial network (GAN) based methods, have demonstrated exceptional performance in image rescaling. However, IRN-based methods tend to produce over-smoothed results, while GAN-based methods easily generate fake details, which thus hinders their real applications. To address this issue, we propose Boundary-aware Decoupled Flow Networks (BDFlow) to generate realistic and visually pleasing results. Unlike previous methods that model high-frequency information as standard Gaussian distribution directly, our BDFlow first decouples the high-frequency information into semantic high-frequency that adheres to a Boundary distribution and non-semantic high-frequency counterpart that adheres to a Gaussian distribution. Specifically, to capture semantic high-frequency parts accurately, we use Boundary-aware Mask (BAM) to constrain the model to produce rich textures, while non-semantic high-frequency part is randomly sampled from a Gaussian distribution. Comprehensive experiments demonstrate that our BDFlow significantly outperforms other state-of-the-art methods while maintaining lower complexity. Notably, our BDFlow improves the PSNR by 4.4 dB and the SSIM by 0.1 on average over GRAIN, utilizing only 74% of the parameters and 20% of the computation. The code will be available at https://github.com/THU-Kingmin/BAFlow.

Jinmin Li, Tao Dai, Yaohua Zha, Yilu Luo, Longfei Lu, Bin Chen, Zhi Wang, Shu-Tao Xia, Jingyun Zhang

Invertible Rescaling Networks (IRNs) and their variants have witnessed remarkable achievements in various image processing tasks like image rescaling. However, we observe that IRNs with deeper networks are difficult to train, thus hindering the representational ability of IRNs. To address this issue, we propose Invertible Residual Rescaling Models (IRRM) for image rescaling by learning a bijection between a high-resolution image and its low-resolution counterpart with a specific distribution. Specifically, we propose IRRM to build a deep network, which contains several Residual Downscaling Modules (RDMs) with long skip connections. Each RDM consists of several Invertible Residual Blocks (IRBs) with short connections. In this way, RDM allows rich low-frequency information to be bypassed by skip connections and forces models to focus on extracting high-frequency information from the image. Extensive experiments show that our IRRM performs significantly better than other state-of-the-art methods with much fewer parameters and complexity. Particularly, our IRRM has respectively PSNR gains of at least 0.3 dB over HCFlow and IRN in the x4 rescaling while only using 60% parameters and 50% FLOPs. The code will be available at https://github.com/THU-Kingmin/IRRM.