Rui Zhao

Xiaoshi Wu, Yiming Hao, Keqiang Sun et al.

Recent text-to-image generative models can generate high-fidelity images from text inputs, but the quality of these generated images cannot be accurately evaluated by existing evaluation metrics. To address this issue, we introduce Human Preference Dataset v2 (HPD v2), a large-scale dataset that captures human preferences on images from a wide range of sources. HPD v2 comprises 798,090 human preference choices on 433,760 pairs of images, making it the largest dataset of its kind. The text prompts and images are deliberately collected to eliminate potential bias, which is a common issue in previous datasets. By fine-tuning CLIP on HPD v2, we obtain Human Preference Score v2 (HPS v2), a scoring model that can more accurately predict human preferences on generated images. Our experiments demonstrate that HPS v2 generalizes better than previous metrics across various image distributions and is responsive to algorithmic improvements of text-to-image generative models, making it a preferable evaluation metric for these models. We also investigate the design of the evaluation prompts for text-to-image generative models, to make the evaluation stable, fair and easy-to-use. Finally, we establish a benchmark for text-to-image generative models using HPS v2, which includes a set of recent text-to-image models from the academic, community and industry. The code and dataset is available at https://github.com/tgxs002/HPSv2 .

Weizhen He, Yiheng Deng, Shixiang Tang et al.

Human intelligence can retrieve any person according to both visual and language descriptions. However, the current computer vision community studies specific person re-identification (ReID) tasks in different scenarios separately, which limits the applications in the real world. This paper strives to resolve this problem by proposing a new instruct-ReID task that requires the model to retrieve images according to the given image or language instructions. Our instruct-ReID is a more general ReID setting, where existing 6 ReID tasks can be viewed as special cases by designing different instructions. We propose a large-scale OmniReID benchmark and an adaptive triplet loss as a baseline method to facilitate research in this new setting. Experimental results show that the proposed multi-purpose ReID model, trained on our OmniReID benchmark without fine-tuning, can improve +0.5%, +0.6%, +7.7% mAP on Market1501, MSMT17, CUHK03 for traditional ReID, +6.4%, +7.1%, +11.2% mAP on PRCC, VC-Clothes, LTCC for clothes-changing ReID, +11.7% mAP on COCAS+ real2 for clothes template based clothes-changing ReID when using only RGB images, +24.9% mAP on COCAS+ real2 for our newly defined language-instructed ReID, +4.3% on LLCM for visible-infrared ReID, +2.6% on CUHK-PEDES for text-to-image ReID. The datasets, the model, and code will be available at https://github.com/hwz-zju/Instruct-ReID.

Shaofeng Zhang, Feng Zhu, Rui Zhao et al.

We propose ADCLR: A ccurate and D ense Contrastive Representation Learning, a novel self-supervised learning framework for learning accurate and dense vision representation. To extract spatial-sensitive information, ADCLR introduces query patches for contrasting in addition with global contrasting. Compared with previous dense contrasting methods, ADCLR mainly enjoys three merits: i) achieving both global-discriminative and spatial-sensitive representation, ii) model-efficient (no extra parameters in addition to the global contrasting baseline), and iii) correspondence-free and thus simpler to implement. Our approach achieves new state-of-the-art performance for contrastive methods. On classification tasks, for ViT-S, ADCLR achieves 77.5% top-1 accuracy on ImageNet with linear probing, outperforming our baseline (DINO) without our devised techniques as plug-in, by 0.5%. For ViT-B, ADCLR achieves 79.8%, 84.0% accuracy on ImageNet by linear probing and finetune, outperforming iBOT by 0.3%, 0.2% accuracy. For dense tasks, on MS-COCO, ADCLR achieves significant improvements of 44.3% AP on object detection, 39.7% AP on instance segmentation, outperforming previous SOTA method SelfPatch by 2.2% and 1.2%, respectively. On ADE20K, ADCLR outperforms SelfPatch by 1.0% mIoU, 1.2% mAcc on the segme

Jingqing Ruan, Yihong Chen, Bin Zhang et al.

With recent advancements in natural language processing, Large Language Models (LLMs) have emerged as powerful tools for various real-world applications. Despite their prowess, the intrinsic generative abilities of LLMs may prove insufficient for handling complex tasks which necessitate a combination of task planning and the usage of external tools. In this paper, we first propose a structured framework tailored for LLM-based AI Agents and discuss the crucial capabilities necessary for tackling intricate problems. Within this framework, we design two distinct types of agents (i.e., one-step agent and sequential agent) to execute the inference process. Subsequently, we instantiate the framework using various LLMs and evaluate their Task Planning and Tool Usage (TPTU) abilities on typical tasks. By highlighting key findings and challenges, our goal is to provide a helpful resource for researchers and practitioners to leverage the power of LLMs in their AI applications. Our study emphasizes the substantial potential of these models, while also identifying areas that need more investigation and improvement.

Bin Zhang, Hangyu Mao, Jingqing Ruan et al.

The remarkable progress in Large Language Models (LLMs) opens up new avenues for addressing planning and decision-making problems in Multi-Agent Systems (MAS). However, as the number of agents increases, the issues of hallucination in LLMs and coordination in MAS have become increasingly prominent. Additionally, the efficient utilization of tokens emerges as a critical consideration when employing LLMs to facilitate the interactions among a substantial number of agents. In this paper, we develop a modular framework called LLaMAC to mitigate these challenges. LLaMAC implements a value distribution encoding similar to that found in the human brain, utilizing internal and external feedback mechanisms to facilitate collaboration and iterative reasoning among its modules. Through evaluations involving system resource allocation and robot grid transportation, we demonstrate the considerable advantages afforded by our proposed approach.

Wenrui Li, Yidan Lu, Yeyu Chai et al.

With the daily influx of 3D data on the internet, text-3D retrieval has gained increasing attention. However, current methods face two major challenges: Hierarchy Representation Collapse (HRC) and Redundancy-Induced Saliency Dilution (RISD). HRC compresses abstract-to-specific and whole-to-part hierarchies in Euclidean embeddings, while RISD averages noisy fragments, obscuring critical semantic cues and diminishing the model's ability to distinguish hard negatives. To address these challenges, we introduce the Hyperbolic Hierarchical Alignment Reasoning Network (H$^{2}$ARN) for text-3D retrieval. H$^{2}$ARN embeds both text and 3D data in a Lorentz-model hyperbolic space, where exponential volume growth inherently preserves hierarchical distances. A hierarchical ordering loss constructs a shrinking entailment cone around each text vector, ensuring that the matched 3D instance falls within the cone, while an instance-level contrastive loss jointly enforces separation from non-matching samples. To tackle RISD, we propose a contribution-aware hyperbolic aggregation module that leverages Lorentzian distance to assess the relevance of each local feature and applies contribution-weighted aggregation guided by hyperbolic geometry, enhancing discriminative regions while suppressing redundancy without additional supervision. We also release the expanded T3DR-HIT v2 benchmark, which contains 8,935 text-to-3D pairs, 2.6 times the original size, covering both fine-grained cultural artefacts and complex indoor scenes. Our codes are available at https://github.com/liwrui/H2ARN.

Yapeng Meng, Songru Yang, Xu Hu et al.

This paper proposes a novel and physically interpretable method for face editing based on arbitrary text prompts. Different from previous GAN-inversion-based face editing methods that manipulate the latent space of GANs, or diffusion-based methods that model image manipulation as a reverse diffusion process, we regard the face editing process as imposing vector flow fields on face images, representing the offset of spatial coordinates and color for each image pixel. Under the above-proposed paradigm, we represent the vector flow field in two ways: 1) explicitly represent the flow vectors with rasterized tensors, and 2) implicitly parameterize the flow vectors as continuous, smooth, and resolution-agnostic neural fields, by leveraging the recent advances of implicit neural representations. The flow vectors are iteratively optimized under the guidance of the pre-trained Contrastive Language-Image Pretraining~(CLIP) model by maximizing the correlation between the edited image and the text prompt. We also propose a learning-based one-shot face editing framework, which is fast and adaptable to any text prompt input. Our method can also be flexibly extended to real-time video face editing. Compared with state-of-the-art text-driven face editing methods, our method can generate physically interpretable face editing results with high identity consistency and image quality. Our code will be made publicly available.

Kaijian Liu, Shixiang Tang, Ziyue Li et al.

Person clustering with multi-modal clues, including faces, bodies, and voices, is critical for various tasks, such as movie parsing and identity-based movie editing. Related methods such as multi-view clustering mainly project multi-modal features into a joint feature space. However, multi-modal clue features are usually rather weakly correlated due to the semantic gap from the modality-specific uniqueness. As a result, these methods are not suitable for person clustering. In this paper, we propose a Relation-Aware Distribution representation Network (RAD-Net) to generate a distribution representation for multi-modal clues. The distribution representation of a clue is a vector consisting of the relation between this clue and all other clues from all modalities, thus being modality agnostic and good for person clustering. Accordingly, we introduce a graph-based method to construct distribution representation and employ a cyclic update policy to refine distribution representation progressively. Our method achieves substantial improvements of +6% and +8.2% in F-score on the Video Person-Clustering Dataset (VPCD) and VoxCeleb2 multi-view clustering dataset, respectively. Codes will be released publicly upon acceptance.

Haoyuan Jiang, Xuantang Xiong, Ziyue Li et al.

Currently, traffic signal control (TSC) methods based on reinforcement learning (RL) have proven superior to traditional methods. However, most RL methods face difficulties when applied in the real world due to three factors: input, output, and the cycle-flow relation. The industry's observable input is much more limited than simulation-based RL methods. For real-world solutions, only flow can be reliably collected, whereas common RL methods need more. For the output action, most RL methods focus on acyclic control, which real-world signal controllers do not support. Most importantly, industry standards require a consistent cycle-flow relationship: non-decreasing and different response strategies for low, medium, and high-level flows, which is ignored by the RL methods. To narrow the gap between RL methods and industry standards, we innovatively propose to use industry solutions to guide the RL agent. Specifically, we design behavior cloning and curriculum learning to guide the agent to mimic and meet industry requirements and, at the same time, leverage the power of exploration and exploitation in RL for better performance. We theoretically prove that such guidance can largely decrease the sample complexity to polynomials in the horizon when searching for an optimal policy. Our rigid experiments show that our method has good cycle-flow relation and superior performance.

Rongyao Fang, Chengqi Duan, Kun Wang et al.

Recent advancements in multimodal foundation models have yielded significant progress in vision-language understanding. Initial attempts have also explored the potential of multimodal large language models (MLLMs) for visual content generation. However, existing works have insufficiently addressed the varying granularity demands of different image generation tasks within a unified MLLM paradigm - from the diversity required in text-to-image generation to the precise controllability needed in image manipulation. In this work, we propose PUMA, emPowering Unified MLLM with Multi-grAnular visual generation. PUMA unifies multi-granular visual features as both inputs and outputs of MLLMs, elegantly addressing the different granularity requirements of various image generation tasks within a unified MLLM framework. Following multimodal pretraining and task-specific instruction tuning, PUMA demonstrates proficiency in a wide range of multimodal tasks. This work represents a significant step towards a truly unified MLLM capable of adapting to the granularity demands of various visual tasks. The code and model will be released in https://github.com/rongyaofang/PUMA.

Jintao Liang, Gang Su, Huifeng Lin et al.

Retrieval-Augmented Generation (RAG) has emerged as a powerful framework to overcome the knowledge limitations of Large Language Models (LLMs) by integrating external retrieval with language generation. While early RAG systems based on static pipelines have shown effectiveness in well-structured tasks, they struggle in real-world scenarios requiring complex reasoning, dynamic retrieval, and multi-modal integration. To address these challenges, the field has shifted toward Reasoning Agentic RAG, a paradigm that embeds decision-making and adaptive tool use directly into the retrieval process. In this paper, we present a comprehensive review of Reasoning Agentic RAG methods, categorizing them into two primary systems: predefined reasoning, which follows fixed modular pipelines to boost reasoning, and agentic reasoning, where the model autonomously orchestrates tool interaction during inference. We analyze representative techniques under both paradigms, covering architectural design, reasoning strategies, and tool coordination. Finally, we discuss key research challenges and propose future directions to advance the flexibility, robustness, and applicability of reasoning agentic RAG systems. Our collection of the relevant research has been organized into a https://github.com/ByebyeMonica/Reasoning-Agentic-RAG.

Chuanghao Ding, Xuejing Liu, Wei Tang et al.

This paper introduces SynthDoc, a novel synthetic document generation pipeline designed to enhance Visual Document Understanding (VDU) by generating high-quality, diverse datasets that include text, images, tables, and charts. Addressing the challenges of data acquisition and the limitations of existing datasets, SynthDoc leverages publicly available corpora and advanced rendering tools to create a comprehensive and versatile dataset. Our experiments, conducted using the Donut model, demonstrate that models trained with SynthDoc's data achieve superior performance in pre-training read tasks and maintain robustness in downstream tasks, despite language inconsistencies. The release of a benchmark dataset comprising 5,000 image-text pairs not only showcases the pipeline's capabilities but also provides a valuable resource for the VDU community to advance research and development in document image recognition. This work significantly contributes to the field by offering a scalable solution to data scarcity and by validating the efficacy of end-to-end models in parsing complex, real-world documents.

Qinghongbing Xie, Zhaoyuan Xia, Feng Zhu et al.

Recently spatial-temporal intelligence of Visual-Language Models (VLMs) has attracted much attention due to its importance for Autonomous Driving, Embodied AI and General Artificial Intelligence. Existing spatial-temporal benchmarks mainly focus on egocentric perspective reasoning with images/video context, or geographic perspective reasoning with graphics context (eg. a map), thus fail to assess VLMs' geographic spatial-temporal intelligence with both images/video and graphics context, which is important for areas like traffic management and emergency response. To address the gaps, we introduce Geo-Temporal Reasoning benchmark (GTR-Bench), a novel challenge for geographic temporal reasoning of moving targets in a large-scale camera network. GTR-Bench is more challenging as it requires multiple perspective switches between maps and videos, joint reasoning across multiple videos with non-overlapping fields of view, and inference over spatial-temporal regions that are unobserved by any video context. Evaluations of more than 10 popular VLMs on GTR-Bench demonstrate that even the best proprietary model, Gemini-2.5-Pro (34.9%), significantly lags behind human performance (78.61%) on geo-temporal reasoning. Moreover, our comprehensive analysis on GTR-Bench reveals three primary deficiencies of current models for geo-temporal reasoning. (1) VLMs' reasoning is impaired by an imbalanced utilization of spatial-temporal context. (2) VLMs are weak in temporal forecasting, which leads to worse performance on temporal-emphasized tasks than on spatial-emphasized tasks. (3) VLMs lack the proficiency to comprehend or align the map data with multi-view video inputs. We believe GTR-Bench offers valuable insights and opens up new opportunities for research and applications in spatial-temporal intelligence. Benchmark and code will be released at https://github.com/X-Luffy/GTR-Bench.

Xuejing Liu, Wei Tang, Jinghui Lu et al.

Recent advancements in multimodal foundation models (e.g., CLIP) have excelled in zero-shot generalization. Prompt tuning involved in the knowledge transfer from foundation models to downstream tasks has gained significant attention recently. Existing prompt-tuning methods in cross-modal learning, however, either solely focus on language branch, or learn vision-language interaction in a shallow mechanism. In this context, we propose a Deeply coupled Cross-modal Prompt learning (DCP) method based on CLIP. DCP flexibly accommodates the interplay between vision and language with a Cross-Modal Prompt Attention (CMPA) mechanism, which enables the mutual exchange of respective representation through a well-connected multi-head attention module progressively and strongly. We then conduct comprehensive few-shot learning experiments on 11 image classification datasets and analyze the robustness to domain shift as well. Thorough experimental analysis evidently demonstrates the superb few-shot generalization and compelling domain adaption capacity of a well-executed DCP. The code can be found at https://github.com/GingL/CMPA.

Bin Zhang, Yuxiao Ye, Guoqing Du et al.

Large Language Models (LLMs) have emerged as a powerful tool in advancing the Text-to-SQL task, significantly outperforming traditional methods. Nevertheless, as a nascent research field, there is still no consensus on the optimal prompt templates and design frameworks. Additionally, existing benchmarks inadequately explore the performance of LLMs across the various sub-tasks of the Text-to-SQL process, which hinders the assessment of LLMs' cognitive capabilities and the optimization of LLM-based solutions. To address the aforementioned issues, we firstly construct a new dataset designed to mitigate the risk of overfitting in LLMs. Then we formulate five evaluation tasks to comprehensively assess the performance of diverse methods across various LLMs throughout the Text-to-SQL process.Our study highlights the performance disparities among LLMs and proposes optimal in-context learning solutions tailored to each task. These findings offer valuable insights for enhancing the development of LLM-based Text-to-SQL systems.

Zhishuai Li, Xiang Wang, Jingjing Zhao et al.

Recent advancements in Text-to-SQL (Text2SQL) emphasize stimulating the large language models (LLM) on in-context learning, achieving significant results. Nevertheless, they face challenges when dealing with verbose database information and complex user intentions. This paper presents a two-stage framework to enhance the performance of current LLM-based natural language to SQL systems. We first introduce a novel prompt representation, called reference-enhanced representation, which includes schema information and randomly sampled cell values from tables to instruct LLMs in generating SQL queries. Then, in the first stage, question-SQL pairs are retrieved as few-shot demonstrations, prompting the LLM to generate a preliminary SQL (PreSQL). After that, the mentioned entities in PreSQL are parsed to conduct schema linking, which can significantly compact the useful information. In the second stage, with the linked schema, we simplify the prompt's schema information and instruct the LLM to produce the final SQL. Finally, as the post-refinement module, we propose using cross-consistency across different LLMs rather than self-consistency within a particular LLM. Our methods achieve new SOTA results on the Spider benchmark, with an execution accuracy of 87.6%.

Sun Yang, Qiong Su, Zhishuai Li et al.

In sophisticated existing Text-to-SQL methods exhibit errors in various proportions, including schema-linking errors (incorrect columns, tables, or extra columns), join errors, nested errors, and group-by errors. Consequently, there is a critical need to filter out unnecessary tables and columns, directing the language models attention to relevant tables and columns with schema-linking, to reduce errors during SQL generation. Previous approaches have involved sorting tables and columns based on their relevance to the question, selecting the top-ranked ones for sorting, or directly identifying the necessary tables and columns for SQL generation. However, these methods face challenges such as lengthy model training times, high consumption of expensive GPT-4 tokens in few-shot prompts, or suboptimal performance in schema linking. Therefore, we propose an inventive schema linking method in two steps: Firstly, generate an initial SQL query by utilizing the complete database schema. Subsequently, extract tables and columns from the initial SQL query to create a concise schema. Using CodeLlama-34B, when comparing the schemas obtained by mainstream methods with ours for SQL generation, our schema performs optimally. Leveraging GPT4, our SQL generation method achieved results that are comparable to mainstream Text-to-SQL methods on the Spider dataset.

Yuke Lou, Yiming Wang, Zhen Wu et al.

Human-object interaction (HOI) synthesis is important for various applications, ranging from virtual reality to robotics. However, acquiring 3D HOI data is challenging due to its complexity and high cost, limiting existing methods to the narrow diversity of object types and interaction patterns in training datasets. This paper proposes a novel zero-shot HOI synthesis framework without relying on end-to-end training on currently limited 3D HOI datasets. The core idea of our method lies in leveraging extensive HOI knowledge from pre-trained Multimodal Models. Given a text description, our system first obtains temporally consistent 2D HOI image sequences using image or video generation models, which are then uplifted to 3D HOI milestones of human and object poses. We employ pre-trained human pose estimation models to extract human poses and introduce a generalizable category-level 6-DoF estimation method to obtain the object poses from 2D HOI images. Our estimation method is adaptive to various object templates obtained from text-to-3D models or online retrieval. A physics-based tracking of the 3D HOI kinematic milestone is further applied to refine both body motions and object poses, yielding more physically plausible HOI generation results. The experimental results demonstrate that our method is capable of generating open-vocabulary HOIs with physical realism and semantic diversity.

Rui Zhao, Beihong Jin, Beibei Li et al.

With the popularity of social media, an increasing number of users are joining group activities on online social platforms. This elicits the requirement of group identification (GI), which is to recommend groups to users. We reveal that users are influenced by both group-level and item-level interests, and these dual-level interests have a collaborative evolution relationship: joining a group expands the user's item interests, further prompting the user to join new groups. Ultimately, the two interests tend to align dynamically. However, existing GI methods fail to fully model this collaborative evolution relationship, ignoring the enhancement of group-level interests on item-level interests, and suffering from false-negative samples when aligning cross-level interests. In order to fully model the collaborative evolution relationship between dual-level user interests, we propose CI4GI, a Collaborative Interest-aware model for Group Identification. Specifically, we design an interest enhancement strategy that identifies additional interests of users from the items interacted with by the groups they have joined as a supplement to item-level interests. In addition, we adopt the distance between interest distributions of two users to optimize the identification of negative samples for a user, mitigating the interference of false-negative samples during cross-level interests alignment. The results of experiments on three real-world datasets demonstrate that CI4GI significantly outperforms state-of-the-art models.

Yuchao Gu, Xintao Wang, Jay Zhangjie Wu et al.

Public large-scale text-to-image diffusion models, such as Stable Diffusion, have gained significant attention from the community. These models can be easily customized for new concepts using low-rank adaptations (LoRAs). However, the utilization of multiple concept LoRAs to jointly support multiple customized concepts presents a challenge. We refer to this scenario as decentralized multi-concept customization, which involves single-client concept tuning and center-node concept fusion. In this paper, we propose a new framework called Mix-of-Show that addresses the challenges of decentralized multi-concept customization, including concept conflicts resulting from existing single-client LoRA tuning and identity loss during model fusion. Mix-of-Show adopts an embedding-decomposed LoRA (ED-LoRA) for single-client tuning and gradient fusion for the center node to preserve the in-domain essence of single concepts and support theoretically limitless concept fusion. Additionally, we introduce regionally controllable sampling, which extends spatially controllable sampling (e.g., ControlNet and T2I-Adaptor) to address attribute binding and missing object problems in multi-concept sampling. Extensive experiments demonstrate that Mix-of-Show is capable of composing multiple customized concepts with high fidelity, including characters, objects, and scenes.

Jianqiu Chen, Zikun Zhou, Mingshan Sun et al.

Many robotics and industry applications have a high demand for the capability to estimate the 6D pose of novel objects from the cluttered scene. However, existing classic pose estimation methods are object-specific, which can only handle the specific objects seen during training. When applied to a novel object, these methods necessitate a cumbersome onboarding process, which involves extensive dataset preparation and model retraining. The extensive duration and resource consumption of onboarding limit their practicality in real-world applications. In this paper, we introduce ZeroPose, a novel zero-shot framework that performs pose estimation following a Discovery-Orientation-Registration (DOR) inference pipeline. This framework generalizes to novel objects without requiring model retraining. Given the CAD model of a novel object, ZeroPose enables in seconds onboarding time to extract visual and geometric embeddings from the CAD model as a prompt. With the prompting of the above embeddings, DOR can discover all related instances and estimate their 6D poses without additional human interaction or presupposing scene conditions. Compared with existing zero-shot methods solved by the render-and-compare paradigm, the DOR pipeline formulates the object pose estimation into a feature-matching problem, which avoids time-consuming online rendering and improves efficiency. Experimental results on the seven datasets show that ZeroPose as a zero-shot method achieves comparable performance with object-specific training methods and outperforms the state-of-the-art zero-shot method with 50x inference speed improvement.

Yizhou Wang, Shixiang Tang, Feng Zhu et al.

The pretrain-finetune paradigm is a classical pipeline in visual learning. Recent progress on unsupervised pretraining methods shows superior transfer performance to their supervised counterparts. This paper revisits this phenomenon and sheds new light on understanding the transferability gap between unsupervised and supervised pretraining from a multilayer perceptron (MLP) perspective. While previous works focus on the effectiveness of MLP on unsupervised image classification where pretraining and evaluation are conducted on the same dataset, we reveal that the MLP projector is also the key factor to better transferability of unsupervised pretraining methods than supervised pretraining methods. Based on this observation, we attempt to close the transferability gap between supervised and unsupervised pretraining by adding an MLP projector before the classifier in supervised pretraining. Our analysis indicates that the MLP projector can help retain intra-class variation of visual features, decrease the feature distribution distance between pretraining and evaluation datasets, and reduce feature redundancy. Extensive experiments on public benchmarks demonstrate that the added MLP projector significantly boosts the transferability of supervised pretraining, e.g. +7.2% top-1 accuracy on the concept generalization task, +5.8% top-1 accuracy for linear evaluation on 12-domain classification tasks, and +0.8% AP on COCO object detection task, making supervised pretraining comparable or even better than unsupervised pretraining.

Zhaowen Li, Zhiyang Chen, Fan Yang et al.

Transformer has been widely used for self-supervised pre-training in Natural Language Processing (NLP) and achieved great success. However, it has not been fully explored in visual self-supervised learning. Meanwhile, previous methods only consider the high-level feature and learning representation from a global perspective, which may fail to transfer to the downstream dense prediction tasks focusing on local features. In this paper, we present a novel Masked Self-supervised Transformer approach named MST, which can explicitly capture the local context of an image while preserving the global semantic information. Specifically, inspired by the Masked Language Modeling (MLM) in NLP, we propose a masked token strategy based on the multi-head self-attention map, which dynamically masks some tokens of local patches without damaging the crucial structure for self-supervised learning. More importantly, the masked tokens together with the remaining tokens are further recovered by a global image decoder, which preserves the spatial information of the image and is more friendly to the downstream dense prediction tasks. The experiments on multiple datasets demonstrate the effectiveness and generality of the proposed method. For instance, MST achieves Top-1 accuracy of 76.9% with DeiT-S only using 300-epoch pre-training by linear evaluation, which outperforms supervised methods with the same epoch by 0.4% and its comparable variant DINO by 1.0\%. For dense prediction tasks, MST also achieves 42.7% mAP on MS COCO object detection and 74.04% mIoU on Cityscapes segmentation only with 100-epoch pre-training.

Zhenghao Chen, Shuhang Gu, Feng Zhu et al.

Exploiting the relationships between attributes is a key challenge for improving multiple facial attribute recognition. In this work, we are concerned with two types of correlations that are spatial and non-spatial relationships. For the spatial correlation, we aggregate attributes with spatial similarity into a part-based group and then introduce a Group Attention Learning to generate the group attention and the part-based group feature. On the other hand, to discover the non-spatial relationship, we model a group-based Graph Correlation Learning to explore affinities of predefined part-based groups. We utilize such affinity information to control the communication between all groups and then refine the learned group features. Overall, we propose a unified network called Multi-scale Group and Graph Network. It incorporates these two newly proposed learning strategies and produces coarse-to-fine graph-based group features for improving facial attribute recognition. Comprehensive experiments demonstrate that our approach outperforms the state-of-the-art methods.

Shijie Yu, Dapeng Chen, Rui Zhao et al.

Person images captured by surveillance cameras are often occluded by various obstacles, which lead to defective feature representation and harm person re-identification (Re-ID) performance. To tackle this challenge, we propose to reconstruct the feature representation of occluded parts by fully exploiting the information of its neighborhood in a gallery image set. Specifically, we first introduce a visible part-based feature by body mask for each person image. Then we identify its neighboring samples using the visible features and reconstruct the representation of the full body by an outlier-removable graph neural network with all the neighboring samples as input. Extensive experiments show that the proposed approach obtains significant improvements. In the large-scale Occluded-DukeMTMC benchmark, our approach achieves 64.2% mAP and 67.6% rank-1 accuracy which outperforms the state-of-the-art approaches by large margins, i.e.,20.4% and 12.5%, respectively, indicating the effectiveness of our method on occluded Re-ID problem.

Yixiao Ge, Xiao Zhang, Ching Lam Choi et al.

The recent studies of knowledge distillation have discovered that ensembling the "dark knowledge" from multiple teachers or students contributes to creating better soft targets for training, but at the cost of significantly more computations and/or parameters. In this work, we present BAtch Knowledge Ensembling (BAKE) to produce refined soft targets for anchor images by propagating and ensembling the knowledge of the other samples in the same mini-batch. Specifically, for each sample of interest, the propagation of knowledge is weighted in accordance with the inter-sample affinities, which are estimated on-the-fly with the current network. The propagated knowledge can then be ensembled to form a better soft target for distillation. In this way, our BAKE framework achieves online knowledge ensembling across multiple samples with only a single network. It requires minimal computational and memory overhead compared to existing knowledge ensembling methods. Extensive experiments demonstrate that the lightweight yet effective BAKE consistently boosts the classification performance of various architectures on multiple datasets, e.g., a significant +0.7% gain of Swin-T on ImageNet with only +1.5% computational overhead and zero additional parameters. BAKE does not only improve the vanilla baselines, but also surpasses the single-network state-of-the-arts on all the benchmarks.

Rui Zhao, Kecheng Zheng, Zheng-Jun Zha et al.

Cross-modal video-text retrieval, a challenging task in the field of vision and language, aims at retrieving corresponding instance giving sample from either modality. Existing approaches for this task all focus on how to design encoding model through a hard negative ranking loss, leaving two key problems unaddressed during this procedure. First, in the training stage, only a mini-batch of instance pairs is available in each iteration. Therefore, this kind of hard negatives is locally mined inside a mini-batch while ignoring the global negative samples among the dataset. Second, there are many text descriptions for one video and each text only describes certain local features of a video. Previous works for this task did not consider to fuse the multiply texts corresponding to a video during the training. In this paper, to solve the above two problems, we propose a novel memory enhanced embedding learning (MEEL) method for videotext retrieval. To be specific, we construct two kinds of memory banks respectively: cross-modal memory module and text center memory module. The cross-modal memory module is employed to record the instance embeddings of all the datasets for global negative mining. To avoid the fast evolving of the embedding in the memory bank during training, we utilize a momentum encoder to update the features by a moving-averaging strategy. The text center memory module is designed to record the center information of the multiple textual instances corresponding to a video, and aims at bridging these textual instances together. Extensive experimental results on two challenging benchmarks, i.e., MSR-VTT and VATEX, demonstrate the effectiveness of the proposed method.

Chen Zhao, Yixiao Ge, Feng Zhu et al.

Correspondence selection aims to correctly select the consistent matches (inliers) from an initial set of putative correspondences. The selection is challenging since putative matches are typically extremely unbalanced, largely dominated by outliers, and the random distribution of such outliers further complicates the learning process for learning-based methods. To address this issue, we propose to progressively prune the correspondences via a local-to-global consensus learning procedure. We introduce a ``pruning'' block that lets us identify reliable candidates among the initial matches according to consensus scores estimated using local-to-global dynamic graphs. We then achieve progressive pruning by stacking multiple pruning blocks sequentially. Our method outperforms state-of-the-arts on robust line fitting, camera pose estimation and retrieval-based image localization benchmarks by significant margins and shows promising generalization ability to different datasets and detector/descriptor combinations.

Zhong Meng, Sarangarajan Parthasarathy, Eric Sun et al.

The external language models (LM) integration remains a challenging task for end-to-end (E2E) automatic speech recognition (ASR) which has no clear division between acoustic and language models. In this work, we propose an internal LM estimation (ILME) method to facilitate a more effective integration of the external LM with all pre-existing E2E models with no additional model training, including the most popular recurrent neural network transducer (RNN-T) and attention-based encoder-decoder (AED) models. Trained with audio-transcript pairs, an E2E model implicitly learns an internal LM that characterizes the training data in the source domain. With ILME, the internal LM scores of an E2E model are estimated and subtracted from the log-linear interpolation between the scores of the E2E model and the external LM. The internal LM scores are approximated as the output of an E2E model when eliminating its acoustic components. ILME can alleviate the domain mismatch between training and testing, or improve the multi-domain E2E ASR. Experimented with 30K-hour trained RNN-T and AED models, ILME achieves up to 15.5% and 6.8% relative word error rate reductions from Shallow Fusion on out-of-domain LibriSpeech and in-domain Microsoft production test sets, respectively.

Yongchao Liu, Yue Jin, Yong Chen et al.

Accelerating deep model training and inference is crucial in practice. Existing deep learning frameworks usually concentrate on optimizing training speed and pay fewer attentions to inference-specific optimizations. Actually, model inference differs from training in terms of computation, e.g. parameters are refreshed each gradient update step during training, but kept invariant during inference. These special characteristics of model inference open new opportunities for its optimization. In this paper, we propose a hardware-aware optimization framework, namely Woodpecker-DL (WPK), to accelerate inference by taking advantage of multiple joint optimizations from the perspectives of graph optimization, automated searches, domain-specific language (DSL) compiler techniques and system-level exploration. In WPK, we investigated two new automated search approaches based on genetic algorithm and reinforcement learning, respectively, to hunt the best operator code configurations targeting specific hardware. A customized DSL compiler is further attached to these search algorithms to generate efficient codes. To create an optimized inference plan, WPK systematically explores high-speed operator implementations from third-party libraries besides our automatically generated codes and singles out the best implementation per operator for use. Extensive experiments demonstrated that on a Tesla P100 GPU, we can achieve the maximum speedup of 5.40 over cuDNN and 1.63 over TVM on individual convolution operators, and run up to 1.18 times faster than TensorRT for end-to-end model inference.

Bo Zhao, Shixiang Tang, Dapeng Chen et al.

With the explosion of digital data in recent years, continuously learning new tasks from a stream of data without forgetting previously acquired knowledge has become increasingly important. In this paper, we propose a new continual learning (CL) setting, namely ``continual representation learning'', which focuses on learning better representation in a continuous way. We also provide two large-scale multi-step benchmarks for biometric identification, where the visual appearance of different classes are highly relevant. In contrast to requiring the model to recognize more learned classes, we aim to learn feature representation that can be better generalized to not only previously unseen images but also unseen classes/identities. For the new setting, we propose a novel approach that performs the knowledge distillation over a large number of identities by applying the neighbourhood selection and consistency relaxation strategies to improve scalability and flexibility of the continual learning model. We demonstrate that existing CL methods can improve the representation in the new setting, and our method achieves better results than the competitors.

Yixiao Ge, Haibo Wang, Feng Zhu et al.

The task of large-scale retrieval-based image localization is to estimate the geographical location of a query image by recognizing its nearest reference images from a city-scale dataset. However, the general public benchmarks only provide noisy GPS labels associated with the training images, which act as weak supervisions for learning image-to-image similarities. Such label noise prevents deep neural networks from learning discriminative features for accurate localization. To tackle this challenge, we propose to self-supervise image-to-region similarities in order to fully explore the potential of difficult positive images alongside their sub-regions. The estimated image-to-region similarities can serve as extra training supervision for improving the network in generations, which could in turn gradually refine the fine-grained similarities to achieve optimal performance. Our proposed self-enhanced image-to-region similarity labels effectively deal with the training bottleneck in the state-of-the-art pipelines without any additional parameters or manual annotations in both training and inference. Our method outperforms state-of-the-arts on the standard localization benchmarks by noticeable margins and shows excellent generalization capability on multiple image retrieval datasets.

Yixiao Ge, Feng Zhu, Dapeng Chen et al.

Domain adaptive object re-ID aims to transfer the learned knowledge from the labeled source domain to the unlabeled target domain to tackle the open-class re-identification problems. Although state-of-the-art pseudo-label-based methods have achieved great success, they did not make full use of all valuable information because of the domain gap and unsatisfying clustering performance. To solve these problems, we propose a novel self-paced contrastive learning framework with hybrid memory. The hybrid memory dynamically generates source-domain class-level, target-domain cluster-level and un-clustered instance-level supervisory signals for learning feature representations. Different from the conventional contrastive learning strategy, the proposed framework jointly distinguishes source-domain classes, and target-domain clusters and un-clustered instances. Most importantly, the proposed self-paced method gradually creates more reliable clusters to refine the hybrid memory and learning targets, and is shown to be the key to our outstanding performance. Our method outperforms state-of-the-arts on multiple domain adaptation tasks of object re-ID and even boosts the performance on the source domain without any extra annotations. Our generalized version on unsupervised object re-ID surpasses state-of-the-art algorithms by considerable 16.7% and 7.9% on Market-1501 and MSMT17 benchmarks.

Shijie Yu, Shihua Li, Dapeng Chen et al.

Recent years have witnessed great progress in person re-identification (re-id). Several academic benchmarks such as Market1501, CUHK03 and DukeMTMC play important roles to promote the re-id research. To our best knowledge, all the existing benchmarks assume the same person will have the same clothes. While in real-world scenarios, it is very often for a person to change clothes. To address the clothes changing person re-id problem, we construct a novel large-scale re-id benchmark named ClOthes ChAnging Person Set (COCAS), which provides multiple images of the same identity with different clothes. COCAS totally contains 62,382 body images from 5,266 persons. Based on COCAS, we introduce a new person re-id setting for clothes changing problem, where the query includes both a clothes template and a person image taking another clothes. Moreover, we propose a two-branch network named Biometric-Clothes Network (BC-Net) which can effectively integrate biometric and clothes feature for re-id under our setting. Experiments show that it is feasible for clothes changing re-id with clothes templates.

Lei Yang, Dapeng Chen, Xiaohang Zhan et al.

Face clustering is an essential tool for exploiting the unlabeled face data, and has a wide range of applications including face annotation and retrieval. Recent works show that supervised clustering can result in noticeable performance gain. However, they usually involve heuristic steps and require numerous overlapped subgraphs, severely restricting their accuracy and efficiency. In this paper, we propose a fully learnable clustering framework without requiring a large number of overlapped subgraphs. Instead, we transform the clustering problem into two sub-problems. Specifically, two graph convolutional networks, named GCN-V and GCN-E, are designed to estimate the confidence of vertices and the connectivity of edges, respectively. With the vertex confidence and edge connectivity, we can naturally organize more relevant vertices on the affinity graph and group them into clusters. Experiments on two large-scale benchmarks show that our method significantly improves clustering accuracy and thus performance of the recognition models trained on top, yet it is an order of magnitude more efficient than existing supervised methods.

Rui Zhao, Xudong Sun, Volker Tresp

In Multi-Goal Reinforcement Learning, an agent learns to achieve multiple goals with a goal-conditioned policy. During learning, the agent first collects the trajectories into a replay buffer, and later these trajectories are selected randomly for replay. However, the achieved goals in the replay buffer are often biased towards the behavior policies. From a Bayesian perspective, when there is no prior knowledge about the target goal distribution, the agent should learn uniformly from diverse achieved goals. Therefore, we first propose a novel multi-goal RL objective based on weighted entropy. This objective encourages the agent to maximize the expected return, as well as to achieve more diverse goals. Secondly, we developed a maximum entropy-based prioritization framework to optimize the proposed objective. For evaluation of this framework, we combine it with Deep Deterministic Policy Gradient, both with or without Hindsight Experience Replay. On a set of multi-goal robotic tasks of OpenAI Gym, we compare our method with other baselines and show promising improvements in both performance and sample-efficiency.

Xiao Zhang, Rui Zhao, Junjie Yan et al.

Cosine-based softmax losses significantly improve the performance of deep face recognition networks. However, these losses always include sensitive hyper-parameters which can make training process unstable, and it is very tricky to set suitable hyper parameters for a specific dataset. This paper addresses this challenge by directly designing the gradients for adaptively training deep neural networks. We first investigate and unify previous cosine softmax losses by analyzing their gradients. This unified view inspires us to propose a novel gradient called P2SGrad (Probability-to-Similarity Gradient), which leverages a cosine similarity instead of classification probability to directly update the testing metrics for updating neural network parameters. P2SGrad is adaptive and hyper-parameter free, which makes the training process more efficient and faster. We evaluate our P2SGrad on three face recognition benchmarks, LFW, MegaFace, and IJB-C. The results show that P2SGrad is stable in training, robust to noise, and achieves state-of-the-art performance on all the three benchmarks.

Xiao Zhang, Rui Zhao, Yu Qiao et al.

The cosine-based softmax losses and their variants achieve great success in deep learning based face recognition. However, hyperparameter settings in these losses have significant influences on the optimization path as well as the final recognition performance. Manually tuning those hyperparameters heavily relies on user experience and requires many training tricks. In this paper, we investigate in depth the effects of two important hyperparameters of cosine-based softmax losses, the scale parameter and angular margin parameter, by analyzing how they modulate the predicted classification probability. Based on these analysis, we propose a novel cosine-based softmax loss, AdaCos, which is hyperparameter-free and leverages an adaptive scale parameter to automatically strengthen the training supervisions during the training process. We apply the proposed AdaCos loss to large-scale face verification and identification datasets, including LFW, MegaFace, and IJB-C 1:1 Verification. Our results show that training deep neural networks with the AdaCos loss is stable and able to achieve high face recognition accuracy. Our method outperforms state-of-the-art softmax losses on all the three datasets.

Jing Xu, Rui Zhao, Feng Zhu et al.

Person re-identification (ReID) is to identify pedestrians observed from different camera views based on visual appearance. It is a challenging task due to large pose variations, complex background clutters and severe occlusions. Recently, human pose estimation by predicting joint locations was largely improved in accuracy. It is reasonable to use pose estimation results for handling pose variations and background clutters, and such attempts have obtained great improvement in ReID performance. However, we argue that the pose information was not well utilized and hasn't yet been fully exploited for person ReID. In this work, we introduce a novel framework called Attention-Aware Compositional Network (AACN) for person ReID. AACN consists of two main components: Pose-guided Part Attention (PPA) and Attention-aware Feature Composition (AFC). PPA is learned and applied to mask out undesirable background features in pedestrian feature maps. Furthermore, pose-guided visibility scores are estimated for body parts to deal with part occlusion in the proposed AFC module. Extensive experiments with ablation analysis show the effectiveness of our method, and state-of-the-art results are achieved on several public datasets, including Market-1501, CUHK03, CUHK01, SenseReID, CUHK03-NP and DukeMTMC-reID.

Hongsheng Li, Rui Zhao, Xiaogang Wang

We present highly efficient algorithms for performing forward and backward propagation of Convolutional Neural Network (CNN) for pixelwise classification on images. For pixelwise classification tasks, such as image segmentation and object detection, surrounding image patches are fed into CNN for predicting the classes of centered pixels via forward propagation and for updating CNN parameters via backward propagation. However, forward and backward propagation was originally designed for whole-image classification. Directly applying it to pixelwise classification in a patch-by-patch scanning manner is extremely inefficient, because surrounding patches of pixels have large overlaps, which lead to a lot of redundant computation. The proposed algorithms eliminate all the redundant computation in convolution and pooling on images by introducing novel d-regularly sparse kernels. It generates exactly the same results as those by patch-by-patch scanning. Convolution and pooling operations with such kernels are able to continuously access memory and can run efficiently on GPUs. A fraction of patches of interest can be chosen from each training image for backward propagation by applying a mask to the error map at the last CNN layer. Its computation complexity is constant with respect to the number of patches sampled from the image. Experiments have shown that our proposed algorithms speed up commonly used patch-by-patch scanning over 1500 times in both forward and backward propagation. The speedup increases with the sizes of images and patches.

Rui Zhao, Wanli Ouyang, Xiaogang Wang

Human eyes can recognize person identities based on small salient regions, i.e. human saliency is distinctive and reliable in pedestrian matching across disjoint camera views. However, such valuable information is often hidden when computing similarities of pedestrian images with existing approaches. Inspired by our user study result of human perception on human saliency, we propose a novel perspective for person re-identification based on learning human saliency and matching saliency distribution. The proposed saliency learning and matching framework consists of four steps: (1) To handle misalignment caused by drastic viewpoint change and pose variations, we apply adjacency constrained patch matching to build dense correspondence between image pairs. (2) We propose two alternative methods, i.e. K-Nearest Neighbors and One-class SVM, to estimate a saliency score for each image patch, through which distinctive features stand out without using identity labels in the training procedure. (3) saliency matching is proposed based on patch matching. Matching patches with inconsistent saliency brings penalty, and images of the same identity are recognized by minimizing the saliency matching cost. (4) Furthermore, saliency matching is tightly integrated with patch matching in a unified structural RankSVM learning framework. The effectiveness of our approach is validated on the VIPeR dataset and the CUHK01 dataset. Our approach outperforms the state-of-the-art person re-identification methods on both datasets.