Erjin Zhou

Zitong Huang, Yiping Bao, Bowen Dong et al.

Weakly-supervised object detection (WSOD) aims to train an object detector only requiring the image-level annotations. Recently, some works have managed to select the accurate boxes generated from a well-trained WSOD network to supervise a semi-supervised detection framework for better performance. However, these approaches simply divide the training set into labeled and unlabeled sets according to the image-level criteria, such that sufficient mislabeled or wrongly localized box predictions are chosen as pseudo ground-truths, resulting in a sub-optimal solution of detection performance. To overcome this issue, we propose a novel WSOD framework with a new paradigm that switches from weak supervision to noisy supervision (W2N). Generally, with given pseudo ground-truths generated from the well-trained WSOD network, we propose a two-module iterative training algorithm to refine pseudo labels and supervise better object detector progressively. In the localization adaptation module, we propose a regularization loss to reduce the proportion of discriminative parts in original pseudo ground-truths, obtaining better pseudo ground-truths for further training. In the semi-supervised module, we propose a two tasks instance-level split method to select high-quality labels for training a semi-supervised detector. Experimental results on different benchmarks verify the effectiveness of W2N, and our W2N outperforms all existing pure WSOD methods and transfer learning methods. Our code is publicly available at https://github.com/1170300714/w2n_wsod.

Delong Chen, Zhao Wu, Fan Liu et al.

Contrastive Language Image Pretraining (CLIP) has received widespread attention, since its learned representations can be transferred well to various downstream tasks. During the training process of the CLIP model, the InfoNCE objective aligns positive image-text pairs and separates negative ones. We show an underlying representation grouping effect during this process: the InfoNCE objective indirectly groups semantically similar representations together via randomly emerged within-modal anchors. Based on this understanding, in this paper, Prototypical Contrastive Language Image Pretraining (ProtoCLIP) is introduced to enhance such grouping by boosting its efficiency and increasing its robustness against the modality gap. Specifically, ProtoCLIP sets up prototype-level discrimination between image and text spaces, which efficiently transfers higher-level structural knowledge. Further, Prototypical Back Translation (PBT) is proposed to decouple representation grouping from representation alignment, resulting in effective learning of meaningful representations under large modality gap. The PBT also enables us to introduce additional external teachers with richer prior language knowledge. ProtoCLIP is trained with an online episodic training strategy, which makes it can be scaled up to unlimited amounts of data. We train our ProtoCLIP on Conceptual Captions and achieved an +5.81% ImageNet linear probing improvement and an +2.01% ImageNet zero-shot classification improvement. On the larger YFCC-15M dataset, ProtoCLIP matches the performance of CLIP with 33% of training time. Codes are available at https://github.com/megvii-research/protoclip.

Wenxuan Guo, Ziyuan Li, Meng Zhang et al.

Vision-Language-Action (VLA) models have shown strong potential for general-purpose robot manipulation by unifying perception and action. However, existing VLA systems primarily rely on textual instructions and struggle to resolve spatial ambiguity in complex scenes with multiple similar objects. To address this limitation, we introduce gesture as a parallel instruction modality and propose a Gesture-aware Vision-Language-Action model (GesVLA). Our approach encodes gesture features directly into the latent space, enabling them to participate in both high-level reasoning and low-level action generation, and adopts a dual-VLM architecture to achieve tight coupling between gesture representations and action policies. At the data level, we construct a scalable gesture data generation pipeline by rendering hand models onto real-world scene images. This reduces the sim-to-real visual gap while producing rich data with diverse motion patterns and corresponding pointing annotations. In addition, we employ a two-stage training strategy to equip the model with both gesture perception and action prediction capabilities. We evaluate our approach on multiple real-world robotic tasks, including a controlled block manipulation task for validation and more practical scenarios such as product and produce selection. Experimental results show that incorporating gesture consistently improves target grounding accuracy and human-robot interaction efficiency, especially in complex and cluttered environments. Project page: https://gwxuan.github.io/GesVLA/.

Zitong Huang, Ze Chen, Zhixing Chen et al.

Few-shot Class-Incremental Learning (FSCIL) aims to continuously learn new classes based on very limited training data without forgetting the old ones encountered. Existing studies solely relied on pure visual networks, while in this paper we solved FSCIL by leveraging the Vision-Language model (e.g., CLIP) and propose a simple yet effective framework, named Learning Prompt with Distribution-based Feature Replay (LP-DiF). We observe that simply using CLIP for zero-shot evaluation can substantially outperform the most influential methods. Then, prompt tuning technique is involved to further improve its adaptation ability, allowing the model to continually capture specific knowledge from each session. To prevent the learnable prompt from forgetting old knowledge in the new session, we propose a pseudo-feature replay approach. Specifically, we preserve the old knowledge of each class by maintaining a feature-level Gaussian distribution with a diagonal covariance matrix, which is estimated by the image features of training images and synthesized features generated from a VAE. When progressing to a new session, pseudo-features are sampled from old-class distributions combined with training images of the current session to optimize the prompt, thus enabling the model to learn new knowledge while retaining old knowledge. Experiments on three prevalent benchmarks, i.e., CIFAR100, mini-ImageNet, CUB-200, and two more challenging benchmarks, i.e., SUN-397 and CUB-200$^*$ proposed in this paper showcase the superiority of LP-DiF, achieving new state-of-the-art (SOTA) in FSCIL. Code is publicly available at https://github.com/1170300714/LP-DiF.

Zitong Huang, Ze Chen, Yuanze Li et al.

Few-Shot Class-Incremental Learning has shown remarkable efficacy in efficient learning new concepts with limited annotations. Nevertheless, the heuristic few-shot annotations may not always cover the most informative samples, which largely restricts the capability of incremental learner. We aim to start from a pool of large-scale unlabeled data and then annotate the most informative samples for incremental learning. Based on this premise, this paper introduces the Active Class-Incremental Learning (ACIL). The objective of ACIL is to select the most informative samples from the unlabeled pool to effectively train an incremental learner, aiming to maximize the performance of the resulting model. Note that vanilla active learning algorithms suffer from class-imbalanced distribution among annotated samples, which restricts the ability of incremental learning. To achieve both class balance and informativeness in chosen samples, we propose Class-Balanced Selection (CBS) strategy. Specifically, we first cluster the features of all unlabeled images into multiple groups. Then for each cluster, we employ greedy selection strategy to ensure that the Gaussian distribution of the sampled features closely matches the Gaussian distribution of all unlabeled features within the cluster. Our CBS can be plugged and played into those CIL methods which are based on pretrained models with prompts tunning technique. Extensive experiments under ACIL protocol across five diverse datasets demonstrate that CBS outperforms both random selection and other SOTA active learning approaches. Code is publicly available at https://github.com/1170300714/CBS.

Zhengxiong Luo, Zhicheng Wang, Yan Huang et al.

Heatmap regression has become the most prevalent choice for nowadays human pose estimation methods. The ground-truth heatmaps are usually constructed via covering all skeletal keypoints by 2D gaussian kernels. The standard deviations of these kernels are fixed. However, for bottom-up methods, which need to handle a large variance of human scales and labeling ambiguities, the current practice seems unreasonable. To better cope with these problems, we propose the scale-adaptive heatmap regression (SAHR) method, which can adaptively adjust the standard deviation for each keypoint. In this way, SAHR is more tolerant of various human scales and labeling ambiguities. However, SAHR may aggravate the imbalance between fore-background samples, which potentially hurts the improvement of SAHR. Thus, we further introduce the weight-adaptive heatmap regression (WAHR) to help balance the fore-background samples. Extensive experiments show that SAHR together with WAHR largely improves the accuracy of bottom-up human pose estimation. As a result, we finally outperform the state-of-the-art model by +1.5AP and achieve 72.0AP on COCO test-dev2017, which is com-arable with the performances of most top-down methods. Source codes are available at https://github.com/greatlog/SWAHR-HumanPose.

Yuanhao Cai, Zhicheng Wang, Zhengxiong Luo et al.

In this paper, we propose a novel method called Residual Steps Network (RSN). RSN aggregates features with the same spatial size (Intra-level features) efficiently to obtain delicate local representations, which retain rich low-level spatial information and result in precise keypoint localization. Additionally, we observe the output features contribute differently to final performance. To tackle this problem, we propose an efficient attention mechanism - Pose Refine Machine (PRM) to make a trade-off between local and global representations in output features and further refine the keypoint locations. Our approach won the 1st place of COCO Keypoint Challenge 2019 and achieves state-of-the-art results on both COCO and MPII benchmarks, without using extra training data and pretrained model. Our single model achieves 78.6 on COCO test-dev, 93.0 on MPII test dataset. Ensembled models achieve 79.2 on COCO test-dev, 77.1 on COCO test-challenge dataset. The source code is publicly available for further research at https://github.com/caiyuanhao1998/RSN/

Hao Shi, Bin Xie, Yingfei Liu et al.

Temporal context is essential for robotic manipulation because such tasks are inherently non-Markovian, yet mainstream VLA models typically overlook it and struggle with long-horizon, temporally dependent tasks. Cognitive science suggests that humans rely on working memory to buffer short-lived representations for immediate control, while the hippocampal system preserves verbatim episodic details and semantic gist of past experience for long-term memory. Inspired by these mechanisms, we propose MemoryVLA, a Cognition-Memory-Action framework for long-horizon robotic manipulation. A pretrained VLM encodes the observation into perceptual and cognitive tokens that form working memory, while a Perceptual-Cognitive Memory Bank stores low-level details and high-level semantics consolidated from it. Working memory retrieves decision-relevant entries from the bank, adaptively fuses them with current tokens, and updates the bank by merging redundancies. Using these tokens, a memory-conditioned diffusion action expert yields temporally aware action sequences. We evaluate MemoryVLA on 150+ simulation and real-world tasks across three robots. On SimplerEnv-Bridge, Fractal, and LIBERO-5 suites, it achieves 71.9%, 72.7%, and 96.5% success rates, respectively, all outperforming state-of-the-art baselines CogACT and pi-0, with a notable +14.6 gain on Bridge. On 12 real-world tasks spanning general skills and long-horizon temporal dependencies, MemoryVLA achieves 84.0% success rate, with long-horizon tasks showing a +26 improvement over state-of-the-art baseline. Project Page: https://shihao1895.github.io/MemoryVLA

Zitong Huang, Ze Chen, Bowen Dong et al.

Model Weight Averaging (MWA) is a technique that seeks to enhance model's performance by averaging the weights of multiple trained models. This paper first empirically finds that 1) the vanilla MWA can benefit the class-imbalanced learning, and 2) performing model averaging in the early epochs of training yields a greater performance improvement than doing that in later epochs. Inspired by these two observations, in this paper we propose a novel MWA technique for class-imbalanced learning tasks named Iterative Model Weight Averaging (IMWA). Specifically, IMWA divides the entire training stage into multiple episodes. Within each episode, multiple models are concurrently trained from the same initialized model weight, and subsequently averaged into a singular model. Then, the weight of this average model serves as a fresh initialization for the ensuing episode, thus establishing an iterative learning paradigm. Compared to vanilla MWA, IMWA achieves higher performance improvements with the same computational cost. Moreover, IMWA can further enhance the performance of those methods employing EMA strategy, demonstrating that IMWA and EMA can complement each other. Extensive experiments on various class-imbalanced learning tasks, i.e., class-imbalanced image classification, semi-supervised class-imbalanced image classification and semi-supervised object detection tasks showcase the effectiveness of our IMWA.

Sen Yang, Zhicheng Wang, Ze Chen et al.

This paper presents a new method to solve keypoint detection and instance association by using Transformer. For bottom-up multi-person pose estimation models, they need to detect keypoints and learn associative information between keypoints. We argue that these problems can be entirely solved by Transformer. Specifically, the self-attention in Transformer measures dependencies between any pair of locations, which can provide association information for keypoints grouping. However, the naive attention patterns are still not subjectively controlled, so there is no guarantee that the keypoints will always attend to the instances to which they belong. To address it we propose a novel approach of supervising self-attention for multi-person keypoint detection and instance association. By using instance masks to supervise self-attention to be instance-aware, we can assign the detected keypoints to their corresponding instances based on the pairwise attention scores, without using pre-defined offset vector fields or embedding like CNN-based bottom-up models. An additional benefit of our method is that the instance segmentation results of any number of people can be directly obtained from the supervised attention matrix, thereby simplifying the pixel assignment pipeline. The experiments on the COCO multi-person keypoint detection challenge and person instance segmentation task demonstrate the effectiveness and simplicity of the proposed method and show a promising way to control self-attention behavior for specific purposes.

Xiaohu Jiang, Ze Chen, Zhicheng Wang et al.

After DETR was proposed, this novel transformer-based detection paradigm which performs several cross-attentions between object queries and feature maps for predictions has subsequently derived a series of transformer-based detection heads. These models iterate object queries after each cross-attention. However, they don't renew the query position which indicates object queries' position information. Thus model needs extra learning to figure out the newest regions that query position should express and need more attention. To fix this issue, we propose the Guided Query Position (GQPos) method to embed the latest location information of object queries to query position iteratively. Another problem of such transformer-based detection heads is the high complexity to perform attention on multi-scale feature maps, which hinders them from improving detection performance at all scales. Therefore we propose a novel fusion scheme named Similar Attention (SiA): besides the feature maps is fused, SiA also fuse the attention weights maps to accelerate the learning of high-resolution attention weight map by well-learned low-resolution attention weight map. Our experiments show that the proposed GQPos improves the performance of a series of models, including DETR, SMCA, YoloS, and HoiTransformer and SiA consistently improve the performance of multi-scale transformer-based detection heads like DETR and HoiTransformer.

Zhengxiong Luo, Zhicheng Wang, Yan Huang et al.

Most existing human pose estimation (HPE) methods exploit multi-scale information by fusing feature maps of four different spatial sizes, \ie $1/4$, $1/8$, $1/16$, and $1/32$ of the input image. There are two drawbacks of this strategy: 1) feature maps of different spatial sizes may be not well aligned spatially, which potentially hurts the accuracy of keypoint location; 2) these scales are fixed and inflexible, which may restrict the generalization ability over various human sizes. Towards these issues, we propose an adaptive dilated convolution (ADC). It can generate and fuse multi-scale features of the same spatial sizes by setting different dilation rates for different channels. More importantly, these dilation rates are generated by a regression module. It enables ADC to adaptively adjust the fused scales and thus ADC may generalize better to various human sizes. ADC can be end-to-end trained and easily plugged into existing methods. Extensive experiments show that ADC can bring consistent improvements to various HPE methods. The source codes will be released for further research.

Yang Hu, Haoxuan You, Zhecan Wang et al.

Graph Neural Network (GNN) has been demonstrated its effectiveness in dealing with non-Euclidean structural data. Both spatial-based and spectral-based GNNs are relying on adjacency matrix to guide message passing among neighbors during feature aggregation. Recent works have mainly focused on powerful message passing modules, however, in this paper, we show that none of the message passing modules is necessary. Instead, we propose a pure multilayer-perceptron-based framework, Graph-MLP with the supervision signal leveraging graph structure, which is sufficient for learning discriminative node representation. In model-level, Graph-MLP only includes multi-layer perceptrons, activation function, and layer normalization. In the loss level, we design a neighboring contrastive (NContrast) loss to bridge the gap between GNNs and MLPs by utilizing the adjacency information implicitly. This design allows our model to be lighter and more robust when facing large-scale graph data and corrupted adjacency information. Extensive experiments prove that even without adjacency information in testing phase, our framework can still reach comparable and even superior performance against the state-of-the-art models in the graph node classification task.

Yanjie Li, Shoukui Zhang, Zhicheng Wang et al.

Human pose estimation deeply relies on visual clues and anatomical constraints between parts to locate keypoints. Most existing CNN-based methods do well in visual representation, however, lacking in the ability to explicitly learn the constraint relationships between keypoints. In this paper, we propose a novel approach based on Token representation for human Pose estimation~(TokenPose). In detail, each keypoint is explicitly embedded as a token to simultaneously learn constraint relationships and appearance cues from images. Extensive experiments show that the small and large TokenPose models are on par with state-of-the-art CNN-based counterparts while being more lightweight. Specifically, our TokenPose-S and TokenPose-L achieve $72.5$ AP and $75.8$ AP on COCO validation dataset respectively, with significant reduction in parameters ($\downarrow80.6\%$; $\downarrow$ $56.8\%$) and GFLOPs ($\downarrow$ $75.3\%$; $\downarrow$ $24.7\%$). Code is publicly available.

Mingyang Shang, Dawei Xiang, Zhicheng Wang et al.

Occlusion is very challenging in pedestrian detection. In this paper, we propose a simple yet effective method named V2F-Net, which explicitly decomposes occluded pedestrian detection into visible region detection and full body estimation. V2F-Net consists of two sub-networks: Visible region Detection Network (VDN) and Full body Estimation Network (FEN). VDN tries to localize visible regions and FEN estimates full-body box on the basis of the visible box. Moreover, to further improve the estimation of full body, we propose a novel Embedding-based Part-aware Module (EPM). By supervising the visibility for each part, the network is encouraged to extract features with essential part information. We experimentally show the effectiveness of V2F-Net by conducting several experiments on two challenging datasets. V2F-Net achieves 5.85% AP gains on CrowdHuman and 2.24% MR-2 improvements on CityPersons compared to FPN baseline. Besides, the consistent gain on both one-stage and two-stage detector validates the generalizability of our method.

Xing Dai, Zeren Jiang, Zhao Wu et al.

In recent years, knowledge distillation has been proved to be an effective solution for model compression. This approach can make lightweight student models acquire the knowledge extracted from cumbersome teacher models. However, previous distillation methods of detection have weak generalization for different detection frameworks and rely heavily on ground truth (GT), ignoring the valuable relation information between instances. Thus, we propose a novel distillation method for detection tasks based on discriminative instances without considering the positive or negative distinguished by GT, which is called general instance distillation (GID). Our approach contains a general instance selection module (GISM) to make full use of feature-based, relation-based and response-based knowledge for distillation. Extensive results demonstrate that the student model achieves significant AP improvement and even outperforms the teacher in various detection frameworks. Specifically, RetinaNet with ResNet-50 achieves 39.1% in mAP with GID on COCO dataset, which surpasses the baseline 36.2% by 2.9%, and even better than the ResNet-101 based teacher model with 38.1% AP.

Zhengxiong Luo, Zhicheng Wang, Yuanhao Cai et al.

In this paper, we propose an efficient human pose estimation network (DANet) by learning deeply aggregated representations. Most existing models explore multi-scale information mainly from features with different spatial sizes. Powerful multi-scale representations usually rely on the cascaded pyramid framework. This framework largely boosts the performance but in the meanwhile makes networks very deep and complex. Instead, we focus on exploiting multi-scale information from layers with different receptive-field sizes and then making full of use this information by improving the fusion method. Specifically, we propose an orthogonal attention block (OAB) and a second-order fusion unit (SFU). The OAB learns multi-scale information from different layers and enhances them by encouraging them to be diverse. The SFU adaptively selects and fuses diverse multi-scale information and suppress the redundant ones. This could maximize the effective information in final fused representations. With the help of OAB and SFU, our single pyramid network may be able to generate deeply aggregated representations that contain even richer multi-scale information and have a larger representing capacity than that of cascaded networks. Thus, our networks could achieve comparable or even better accuracy with much smaller model complexity. Specifically, our \mbox{DANet-72} achieves $70.5$ in AP score on COCO test-dev set with only $1.0G$ FLOPs. Its speed on a CPU platform achieves $58$ Persons-Per-Second~(PPS).

Ling Yang, Liangliang Li, Zilun Zhang et al.

Most graph-network-based meta-learning approaches model instance-level relation of examples. We extend this idea further to explicitly model the distribution-level relation of one example to all other examples in a 1-vs-N manner. We propose a novel approach named distribution propagation graph network (DPGN) for few-shot learning. It conveys both the distribution-level relations and instance-level relations in each few-shot learning task. To combine the distribution-level relations and instance-level relations for all examples, we construct a dual complete graph network which consists of a point graph and a distribution graph with each node standing for an example. Equipped with dual graph architecture, DPGN propagates label information from labeled examples to unlabeled examples within several update generations. In extensive experiments on few-shot learning benchmarks, DPGN outperforms state-of-the-art results by a large margin in 5% $\sim$ 12% under supervised setting and 7% $\sim$ 13% under semi-supervised setting. Code will be released.

Guan'an Wang, Shuo Yang, Huanyu Liu et al.

Occluded person re-identification (ReID) aims to match occluded person images to holistic ones across dis-joint cameras. In this paper, we propose a novel framework by learning high-order relation and topology information for discriminative features and robust alignment. At first, we use a CNN backbone and a key-points estimation model to extract semantic local features. Even so, occluded images still suffer from occlusion and outliers. Then, we view the local features of an image as nodes of a graph and propose an adaptive direction graph convolutional (ADGC)layer to pass relation information between nodes. The proposed ADGC layer can automatically suppress the message-passing of meaningless features by dynamically learning di-rection and degree of linkage. When aligning two groups of local features from two images, we view it as a graph matching problem and propose a cross-graph embedded-alignment (CGEA) layer to jointly learn and embed topology information to local features, and straightly predict similarity score. The proposed CGEA layer not only take full use of alignment learned by graph matching but also re-place sensitive one-to-one matching with a robust soft one. Finally, extensive experiments on occluded, partial, and holistic ReID tasks show the effectiveness of our proposed method. Specifically, our framework significantly outperforms state-of-the-art by6.5%mAP scores on Occluded-Duke dataset.

Erjin Zhou, Zhimin Cao, Jian Sun

In this paper, we propose a method, called GridFace, to reduce facial geometric variations and improve the recognition performance. Our method rectifies the face by local homography transformations, which are estimated by a face rectification network. To encourage the image generation with canonical views, we apply a regularization based on the natural face distribution. We learn the rectification network and recognition network in an end-to-end manner. Extensive experiments show our method greatly reduces geometric variations, and gains significant improvements in unconstrained face recognition scenarios.

Zhiao Huang, Erjin Zhou, Zhimin Cao

Facial landmark localization plays an important role in face recognition and analysis applications. In this paper, we give a brief introduction to a coarse-to-fine pipeline with neural networks and sequential regression. First, a global convolutional network is applied to the holistic facial image to give an initial landmark prediction. A pyramid of multi-scale local image patches is then cropped to feed to a new network for each landmark to refine the prediction. As the refinement network outputs a more accurate position estimation than the input, such procedure could be repeated several times until the estimation converges. We evaluate our system on the 300-W dataset [11] and it outperforms the recent state-of-the-arts.

Erjin Zhou, Zhimin Cao, Qi Yin

Face recognition performance improves rapidly with the recent deep learning technique developing and underlying large training dataset accumulating. In this paper, we report our observations on how big data impacts the recognition performance. According to these observations, we build our Megvii Face Recognition System, which achieves 99.50% accuracy on the LFW benchmark, outperforming the previous state-of-the-art. Furthermore, we report the performance in a real-world security certification scenario. There still exists a clear gap between machine recognition and human performance. We summarize our experiments and present three challenges lying ahead in recent face recognition. And we indicate several possible solutions towards these challenges. We hope our work will stimulate the community's discussion of the difference between research benchmark and real-world applications.