Jianqin Yin

Anthony Cioppa, Silvio Giancola, Vladimir Somers et al. · pku

The SoccerNet 2023 challenges were the third annual video understanding challenges organized by the SoccerNet team. For this third edition, the challenges were composed of seven vision-based tasks split into three main themes. The first theme, broadcast video understanding, is composed of three high-level tasks related to describing events occurring in the video broadcasts: (1) action spotting, focusing on retrieving all timestamps related to global actions in soccer, (2) ball action spotting, focusing on retrieving all timestamps related to the soccer ball change of state, and (3) dense video captioning, focusing on describing the broadcast with natural language and anchored timestamps. The second theme, field understanding, relates to the single task of (4) camera calibration, focusing on retrieving the intrinsic and extrinsic camera parameters from images. The third and last theme, player understanding, is composed of three low-level tasks related to extracting information about the players: (5) re-identification, focusing on retrieving the same players across multiple views, (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams, and (7) jersey number recognition, focusing on recognizing the jersey number of players from tracklets. Compared to the previous editions of the SoccerNet challenges, tasks (2-3-7) are novel, including new annotations and data, task (4) was enhanced with more data and annotations, and task (6) now focuses on end-to-end approaches. More information on the tasks, challenges, and leaderboards are available on https://www.soccer-net.org. Baselines and development kits can be found on https://github.com/SoccerNet.

Wei Dai, Rui Liu, Tianyi Wu et al.

Accurate and unbiased examinations of skin lesions are critical for the early diagnosis and treatment of skin diseases. Visual features of skin lesions vary significantly because the images are collected from patients with different lesion colours and morphologies by using dissimilar imaging equipment. Recent studies have reported that ensembled convolutional neural networks (CNNs) are practical to classify the images for early diagnosis of skin disorders. However, the practical use of these ensembled CNNs is limited as these networks are heavyweight and inadequate for processing contextual information. Although lightweight networks (e.g., MobileNetV3 and EfficientNet) were developed to achieve parameters reduction for implementing deep neural networks on mobile devices, insufficient depth of feature representation restricts the performance. To address the existing limitations, we develop a new lite and effective neural network, namely HierAttn. The HierAttn applies a novel deep supervision strategy to learn the local and global features by using multi-stage and multi-branch attention mechanisms with only one training loss. The efficacy of HierAttn was evaluated by using the dermoscopy images dataset ISIC2019 and smartphone photos dataset PAD-UFES-20 (PAD2020). The experimental results show that HierAttn achieves the best accuracy and area under the curve (AUC) among the state-of-the-art lightweight networks. The code is available at https://github.com/anthonyweidai/HierAttn.

Shaojie Zhang, Jianqin Yin, Yonghao Dang et al.

Graph convolution networks (GCNs) have achieved remarkable performance in skeleton-based action recognition. However, previous GCN-based methods rely on elaborate human priors excessively and construct complex feature aggregation mechanisms, which limits the generalizability and effectiveness of networks. To solve these problems, we propose a novel Spatial Topology Gating Unit (STGU), an MLP-based variant without extra priors, to capture the co-occurrence topology features that encode the spatial dependency across all joints. In STGU, to learn the point-wise topology features, a new gate-based feature interaction mechanism is introduced to activate the features point-to-point by the attention map generated from the input sample. Based on the STGU, we propose the first MLP-based model, SiT-MLP, for skeleton-based action recognition in this work. Compared with previous methods on three large-scale datasets, SiT-MLP achieves competitive performance. In addition, SiT-MLP reduces the parameters significantly with favorable results. The code will be available at https://github.com/BUPTSJZhang/SiT?MLP.

Yuanyuan Jiang, Jianqin Yin, Yonghao Dang

Audio-visual event (AVE) localization has attracted much attention in recent years. Most existing methods are often limited to independently encoding and classifying each video segment separated from the full video (which can be regarded as the segment-level representations of events). However, they ignore the semantic consistency of the event within the same full video (which can be considered as the video-level representations of events). In contrast to existing methods, we propose a novel video-level semantic consistency guidance network for the AVE localization task. Specifically, we propose an event semantic consistency modeling (ESCM) module to explore video-level semantic information for semantic consistency modeling. It consists of two components: a cross-modal event representation extractor (CERE) and an intra-modal semantic consistency enhancer (ISCE). CERE is proposed to obtain the event semantic information at the video level. Furthermore, ISCE takes video-level event semantics as prior knowledge to guide the model to focus on the semantic continuity of an event within each modality. Moreover, we propose a new negative pair filter loss to encourage the network to filter out the irrelevant segment pairs and a new smooth loss to further increase the gap between different categories of events in the weakly-supervised setting. We perform extensive experiments on the public AVE dataset and outperform the state-of-the-art methods in both fully- and weakly-supervised settings, thus verifying the effectiveness of our method.The code is available at https://github.com/Bravo5542/VSCG.

Yonghao Dang, Jianqin Yin, Shaojie Zhang et al.

Estimating human poses from videos is critical in human-computer interaction. Joints cooperate rather than move independently during human movement. There are both spatial and temporal correlations between joints. Despite the positive results of previous approaches, most focus on modeling the spatial correlation between joints while only straightforwardly integrating features along the temporal dimension, ignoring the temporal correlation between joints. In this work, we propose a plug-and-play kinematics modeling module (KMM) to explicitly model temporal correlations between joints across different frames by calculating their temporal similarity. In this way, KMM can capture motion cues of the current joint relative to all joints in different time. Besides, we formulate video-based human pose estimation as a Markov Decision Process and design a novel kinematics modeling network (KIMNet) to simulate the Markov Chain, allowing KIMNet to locate joints recursively. Our approach achieves state-of-the-art results on two challenging benchmarks. In particular, KIMNet shows robustness to the occlusion. The code will be released at https://github.com/YHDang/KIMNet.

Meng Chen, Jiawei Tu, Chao Qi et al.

The significant advancements in embodied vision navigation have raised concerns about its susceptibility to adversarial attacks exploiting deep neural networks. Investigating the adversarial robustness of embodied vision navigation is crucial, especially given the threat of 3D physical attacks that could pose risks to human safety. However, existing attack methods for embodied vision navigation often lack physical feasibility due to challenges in transferring digital perturbations into the physical world. Moreover, current physical attacks for object detection struggle to achieve both multi-view effectiveness and visual naturalness in navigation scenarios. To address this, we propose a practical attack method for embodied navigation by attaching adversarial patches to objects, where both opacity and textures are learnable. Specifically, to ensure effectiveness across varying viewpoints, we employ a multi-view optimization strategy based on object-aware sampling, which optimizes the patch's texture based on feedback from the vision-based perception model used in navigation. To make the patch inconspicuous to human observers, we introduce a two-stage opacity optimization mechanism, in which opacity is fine-tuned after texture optimization. Experimental results demonstrate that our adversarial patches decrease the navigation success rate by an average of 22.39%, outperforming previous methods in practicality, effectiveness, and naturalness. Code is available at: https://github.com/chen37058/Physical-Attacks-in-Embodied-Nav

Jiajun Fu, Fuxing Yang, Yonghao Dang et al.

Human motion prediction is challenging due to the complex spatiotemporal feature modeling. Among all methods, graph convolution networks (GCNs) are extensively utilized because of their superiority in explicit connection modeling. Within a GCN, the graph correlation adjacency matrix drives feature aggregation and is the key to extracting predictive motion features. State-of-the-art methods decompose the spatiotemporal correlation into spatial correlations for each frame and temporal correlations for each joint. Directly parameterizing these correlations introduces redundant parameters to represent common relations shared by all frames and all joints. Besides, the spatiotemporal graph adjacency matrix is the same for different motion samples and cannot reflect sample-wise correspondence variances. To overcome these two bottlenecks, we propose dynamic spatiotemporal decompose GC (DSTD-GC), which only takes 28.6% parameters of the state-of-the-art GC. The key of DSTD-GC is constrained dynamic correlation modeling, which explicitly parameterizes the common static constraints as a spatial/temporal vanilla adjacency matrix shared by all frames/joints and dynamically extracts correspondence variances for each frame/joint with an adjustment modeling function. For each sample, the common constrained adjacency matrices are fixed to represent generic motion patterns, while the extracted variances complete the matrices with specific pattern adjustments. Meanwhile, we mathematically reformulate GCs on spatiotemporal graphs into a unified form and find that DSTD-GC relaxes certain constraints of other GC, which contributes to a better representation capability. By combining DSTD-GC with prior knowledge, we propose a powerful spatiotemporal GCN called DSTD-GCN, which outperforms SOTA methods by $3.9\% \sim 8.7\%$ in prediction accuracy with $55.0\% \sim 96.9\%$ fewer parameters.

Zehao Wang, Haobo Yue, Zhicheng Zhang et al.

Sound Event Detection (SED) plays a vital role in comprehending and perceiving acoustic scenes. Previous methods have demonstrated impressive capabilities. However, they are deficient in learning features of complex scenes from heterogeneous dataset. In this paper, we introduce a novel dual-branch architecture named Mutual-Assistance Tuning and Dual-Branch Aggregating for Heterogeneous Sound Event Detection (MTDA-HSED). The MTDA-HSED architecture employs the Mutual-Assistance Audio Adapter (M3A) to effectively tackle the multi-scenario problem and uses the Dual-Branch Mid-Fusion (DBMF) module to tackle the multi-granularity problem. Specifically, M3A is integrated into the BEATs block as an adapter to improve the BEATs' performance by fine-tuning it on the multi-scenario dataset. The DBMF module connects BEATs and CNN branches, which facilitates the deep fusion of information from the BEATs and the CNN branches. Experimental results show that the proposed methods exceed the baseline of mpAUC by \textbf{$5\%$} on the DESED and MAESTRO Real datasets. Code is available at https://github.com/Visitor-W/MTDA.

Chao Qi, Jianqin Yin, Jinghang Xu et al.

This work introduces a new task of instance-incremental scene graph generation: Given a scene of the point cloud, representing it as a graph and automatically increasing novel instances. A graph denoting the object layout of the scene is finally generated. It is an important task since it helps to guide the insertion of novel 3D objects into a real-world scene in vision-based applications like augmented reality. It is also challenging because the complexity of the real-world point cloud brings difficulties in learning object layout experiences from the observation data (non-empty rooms with labeled semantics). We model this task as a conditional generation problem and propose a 3D autoregressive framework based on normalizing flows (3D-ANF) to address it. First, we represent the point cloud as a graph by extracting the label semantics and contextual relationships. Next, a model based on normalizing flows is introduced to map the conditional generation of graphic elements into the Gaussian process. The mapping is invertible. Thus, the real-world experiences represented in the observation data can be modeled in the training phase, and novel instances can be autoregressively generated based on the Gaussian process in the testing phase. To evaluate the performance of our method sufficiently, we implement this new task on the indoor benchmark dataset 3DSSG-O27R16 and our newly proposed graphical dataset of outdoor scenes GPL3D. Experiments show that our method generates reliable novel graphs from the real-world point cloud and achieves state-of-the-art performance on the datasets.

Tingxiu Chen, Jianqin Yin, Jin Tang

In recent years, audio-visual event localization has attracted much attention. It's purpose is to detect the segment containing audio-visual events and recognize the event category from untrimmed videos. Existing methods use audio-guided visual attention to lead the model pay attention to the spatial area of the ongoing event, devoting to the correlation between audio and visual information but ignoring the correlation between audio and spatial motion. We propose a past and future motion extraction (pf-ME) module to mine the visual motion from videos ,embedded into the past and future motion guided network (PFAGN), and motion guided audio attention (MGAA) module to achieve focusing on the information related to interesting events in audio modality through the past and future visual motion. We choose AVE as the experimental verification dataset and the experiments show that our method outperforms the state-of-the-arts in both supervised and weakly-supervised settings.

Binglu Ren, Jianqin Yin

In the perception task of autonomous driving, multi-modal methods have become a trend due to the complementary characteristics of LiDAR point clouds and image data. However, the performance of multi-modal methods is usually limited by the sparsity of the point cloud or the noise problem caused by the misalignment between LiDAR and the camera. To solve these two problems, we present a new concept, Voxel Region (VR), which is obtained by projecting the sparse local point clouds in each voxel dynamically. And we propose a novel fusion method named Sparse-to-Dense Voxel Region Fusion (SDVRF). Specifically, more pixels of the image feature map inside the VR are gathered to supplement the voxel feature extracted from sparse points and achieve denser fusion. Meanwhile, different from prior methods, which project the size-fixed grids, our strategy of generating dynamic regions achieves better alignment and avoids introducing too much background noise. Furthermore, we propose a multi-scale fusion framework to extract more contextual information and capture the features of objects of different sizes. Experiments on the KITTI dataset show that our method improves the performance of different baselines, especially on classes of small size, including Pedestrian and Cyclist.

Guoliang Xu, Jianqin Yin

The task of Group Activity Recognition (GAR) aims to predict the activity category of the group by learning the actor spatial-temporal interaction relation in the group. Therefore, an effective actor relation learning method is crucial for the GAR task. The previous works mainly learn the interaction relation by the well-designed GCNs or Transformers. For example, to infer the actor interaction relation, GCNs need a learnable adjacency, and Transformers need to calculate the self-attention. Although the above methods can model the interaction relation effectively, they also increase the complexity of the model (the number of parameters and computations). In this paper, we design a novel MLP-based method for Actor Interaction Relation learning (MLP-AIR) in GAR. Compared with GCNs and Transformers, our method has a competitive but conceptually and technically simple alternative, significantly reducing the complexity. Specifically, MLP-AIR includes three sub-modules: MLP-based Spatial relation modeling module (MLP-S), MLP-based Temporal relation modeling module (MLP-T), and MLP-based Relation refining module (MLP-R). MLP-S is used to model the spatial relation between different actors in each frame. MLP-T is used to model the temporal relation between different frames for each actor. MLP-R is used further to refine the relation between different dimensions of relation features to improve the feature's expression ability. To evaluate the MLP-AIR, we conduct extensive experiments on two widely used benchmarks, including the Volleyball and Collective Activity datasets. Experimental results demonstrate that MLP-AIR can get competitive results but with low complexity.

Jiajun Fu, Yonghao Dang, Ruoqi Yin et al.

This technical report describes our first-place solution to the pose estimation challenge at ECCV 2022 Visual Perception for Navigation in Human Environments Workshop. In this challenge, we aim to estimate human poses from in-the-wild stitched panoramic images. Our method is built based on Faster R-CNN for human detection, and HRNet for human pose estimation. We describe technical details for the JRDB-Pose dataset, together with some experimental results. In the competition, we achieved 0.303 $\text{OSPA}_{\text{IOU}}$ and 64.047\% $\text{AP}_{\text{0.5}}$ on the test set of JRDB-Pose.

Xiaofa Liu, Jianqin Yin, Yuan Sun et al.

In this paper, we develop an efficient multi-scale network to predict action classes in partial videos in an end-to-end manner. Unlike most existing methods with offline feature generation, our method directly takes frames as input and further models motion evolution on two different temporal scales.Therefore, we solve the complexity problems of the two stages of modeling and the problem of insufficient temporal and spatial information of a single scale. Our proposed End-to-End MultiScale Network (E2EMSNet) is composed of two scales which are named segment scale and observed global scale. The segment scale leverages temporal difference over consecutive frames for finer motion patterns by supplying 2D convolutions. For observed global scale, a Long Short-Term Memory (LSTM) is incorporated to capture motion features of observed frames. Our model provides a simple and efficient modeling framework with a small computational cost. Our E2EMSNet is evaluated on three challenging datasets: BIT, HMDB51, and UCF101. The extensive experiments demonstrate the effectiveness of our method for action prediction in videos.

Da Mu, Zhicheng Zhang, Haobo Yue et al.

In the Sound Event Localization and Detection (SELD) task, Transformer-based models have demonstrated impressive capabilities. However, the quadratic complexity of the Transformer's self-attention mechanism results in computational inefficiencies. In this paper, we propose a network architecture for SELD called SELD-Mamba, which utilizes Mamba, a selective state-space model. We adopt the Event-Independent Network V2 (EINV2) as the foundational framework and replace its Conformer blocks with bidirectional Mamba blocks to capture a broader range of contextual information while maintaining computational efficiency. Additionally, we implement a two-stage training method, with the first stage focusing on Sound Event Detection (SED) and Direction of Arrival (DoA) estimation losses, and the second stage reintroducing the Source Distance Estimation (SDE) loss. Our experimental results on the 2024 DCASE Challenge Task3 dataset demonstrate the effectiveness of the selective state-space model in SELD and highlight the benefits of the two-stage training approach in enhancing SELD performance.

Ruoqi Yin, Jianqin Yin

In this paper, we concern on the bottom-up paradigm in multi-person pose estimation (MPPE). Most previous bottom-up methods try to consider the relation of instances to identify different body parts during the post processing, while ignoring to model the relation among instances or environment in the feature learning process. In addition, most existing works adopt the operations of upsampling and downsampling. During the sampling process, there will be a problem of misalignment with the source features, resulting in deviations in the keypoint features learned by the model. To overcome the above limitations, we propose a convolutional neural network for bottom-up human pose estimation. It invovles two basic modules: (i) Global Relation Modeling (GRM) module globally learns relation (e.g., environment context, instance interactive information) among region of image by fusing multiple stages features in the feature learning process. It combines with the spatial-channel attention mechanism, which focuses on achieving adaptability in spatial and channel dimensions. (ii) Multi-branch Feature Align (MFA) module aggregates features from multiple branches to align fused feature and obtain refined local keypoint representation. Our model has the ability to focus on different granularity from local to global regions, which significantly boosts the performance of the multi-person pose estimation. Our results on the COCO and CrowdPose datasets demonstrate that it is an efficient framework for multi-person pose estimation.

Yonghao Dang, Jianqin Yin, Liyuan Liu et al.

Multi-person pose estimation (MPPE) presents a formidable yet crucial challenge in computer vision. Most existing methods predominantly concentrate on isolated interaction either between instances or joints, which is inadequate for scenarios demanding concurrent localization of both instances and joints. This paper introduces a novel CNN-based single-stage method, named Dual-path Hierarchical Relation Network (DHRNet), to extract instance-to-joint and joint-to-instance interactions concurrently. Specifically, we design a dual-path interaction modeling module (DIM) that strategically organizes cross-instance and cross-joint interaction modeling modules in two complementary orders, enriching interaction information by integrating merits from different correlation modeling branches. Notably, DHRNet excels in joint localization by leveraging information from other instances and joints. Extensive evaluations on challenging datasets, including COCO, CrowdPose, and OCHuman datasets, showcase DHRNet's state-of-the-art performance. The code will be released at https://github.com/YHDang/dhrnet-multi-pose-estimation.

Guoliang Xu, Jianqin Yin, Feng Zhou et al.

Previous methods usually only extract the image modality's information to recognize group activity. However, mining image information is approaching saturation, making it difficult to extract richer information. Therefore, extracting complementary information from other modalities to supplement image information has become increasingly important. In fact, action labels provide clear text information to express the action's semantics, which existing methods often overlook. Thus, we propose ActivityCLIP, a plug-and-play method for mining the text information contained in the action labels to supplement the image information for enhancing group activity recognition. ActivityCLIP consists of text and image branches, where the text branch is plugged into the image branch (The off-the-shelf image-based method). The text branch includes Image2Text and relation modeling modules. Specifically, we propose the knowledge transfer module, Image2Text, which adapts image information into text information extracted by CLIP via knowledge distillation. Further, to keep our method convenient, we add fewer trainable parameters based on the relation module of the image branch to model interaction relation in the text branch. To show our method's generality, we replicate three representative methods by ActivityCLIP, which adds only limited trainable parameters, achieving favorable performance improvements for each method. We also conduct extensive ablation studies and compare our method with state-of-the-art methods to demonstrate the effectiveness of ActivityCLIP.

Chengxu Duan, Zhicheng Zhang, Xiaoli Liu et al.

Human motion prediction has achieved a brilliant performance with the help of convolution-based neural networks. However, currently, there is no work evaluating the potential risk in human motion prediction when facing adversarial attacks. The adversarial attack will encounter problems against human motion prediction in naturalness and data scale. To solve the problems above, we propose a new adversarial attack method that generates the worst-case perturbation by maximizing the human motion predictor's prediction error with physical constraints. Specifically, we introduce a novel adaptable scheme that facilitates the attack to suit the scale of the target pose and two physical constraints to enhance the naturalness of the adversarial example. The evaluating experiments on three datasets show that the prediction errors of all target models are enlarged significantly, which means current convolution-based human motion prediction models are vulnerable to the proposed attack. Based on the experimental results, we provide insights on how to enhance the adversarial robustness of the human motion predictor and how to improve the adversarial attack against human motion prediction.

Lizhi Wang, Feng Zhou, Bo yu et al.

Recent advancements in 3D reconstruction technologies have paved the way for high-quality and real-time rendering of complex 3D scenes. Despite these achievements, a notable challenge persists: it is difficult to precisely reconstruct specific objects from large scenes. Current scene reconstruction techniques frequently result in the loss of object detail textures and are unable to reconstruct object portions that are occluded or unseen in views. To address this challenge, we delve into the meticulous 3D reconstruction of specific objects within large scenes and propose a framework termed OMEGAS: Object Mesh Extraction from Large Scenes Guided by Gaussian Segmentation. Specifically, we proposed a novel 3D target segmentation technique based on 2D Gaussian Splatting, which segments 3D consistent target masks in multi-view scene images and generates a preliminary target model. Moreover, to reconstruct the unseen portions of the target, we propose a novel target replenishment technique driven by large-scale generative diffusion priors. We demonstrate that our method can accurately reconstruct specific targets from large scenes, both quantitatively and qualitatively. Our experiments show that OMEGAS significantly outperforms existing reconstruction methods across various scenarios. Our project page is at: https://github.com/CrystalWlz/OMEGAS

Zhicheng Zhang, Xueyao Sun, Yonghao Dang et al.

Human Pose Estimation (HPE) plays a crucial role in computer vision applications. However, it is difficult to deploy state-of-the-art models on resouce-limited devices due to the high computational costs of the networks. In this work, a binary human pose estimator named BiHRNet(Binary HRNet) is proposed, whose weights and activations are expressed as $\pm$1. BiHRNet retains the keypoint extraction ability of HRNet, while using fewer computing resources by adapting binary neural network (BNN). In order to reduce the accuracy drop caused by network binarization, two categories of techniques are proposed in this work. For optimizing the training process for binary pose estimator, we propose a new loss function combining KL divergence loss with AWing loss, which makes the binary network obtain more comprehensive output distribution from its real-valued counterpart to reduce information loss caused by binarization. For designing more binarization-friendly structures, we propose a new information reconstruction bottleneck called IR Bottleneck to retain more information in the initial stage of the network. In addition, we also propose a multi-scale basic block called MS-Block for information retention. Our work has less computation cost with few precision drop. Experimental results demonstrate that BiHRNet achieves a PCKh of 87.9 on the MPII dataset, which outperforms all binary pose estimation networks. On the challenging of COCO dataset, the proposed method enables the binary neural network to achieve 70.8 mAP, which is better than most tested lightweight full-precision networks.

Yiyang Ma, Feng Zhou, Xuedan Yin et al.

Leveraging pre-trained Diffusion Transformers (DiTs) for high-resolution (HR) image synthesis often leads to spatial layout collapse and degraded texture fidelity. Prior work mitigates these issues with complex pipelines that first perform a base-resolution (i.e., training-resolution) denoising process to guide HR generation. We instead explore the intrinsic generative mechanisms of DiTs and propose ResDiT, a training-free method that scales resolution efficiently. We identify the core factor governing spatial layout, position embeddings (PEs), and show that the original PEs encode incorrect positional information when extrapolated to HR, which triggers layout collapse. To address this, we introduce a PE scaling technique that rectifies positional encoding under resolution changes. To further remedy low-fidelity details, we develop a local-enhancement mechanism grounded in base-resolution local attention. We design a patch-level fusion module that aggregates global and local cues, together with a Gaussian-weighted splicing strategy that eliminates grid artifacts. Comprehensive evaluations demonstrate that ResDiT consistently delivers high-fidelity, high-resolution image synthesis and integrates seamlessly with downstream tasks, including spatially controlled generation.

Feng Zhou, Wenkai Guo, Pu Cao et al.

Sparse-view 3D Gaussian Splatting (3DGS) often overfits to the training views, leading to artifacts like blurring in novel view rendering. Prior work addresses it either by enhancing the initialization (\emph{i.e.}, the point cloud from Structure-from-Motion (SfM)) or by adding training-time constraints (regularization) to the 3DGS optimization. Yet our controlled ablations reveal that initialization is the decisive factor: it determines the attainable performance band in sparse-view 3DGS, while training-time constraints yield only modest within-band improvements at extra cost. Given initialization's primacy, we focus our design there. Although SfM performs poorly under sparse views due to its reliance on feature matching, it still provides reliable seed points. Thus, building on SfM, our effort aims to supplement the regions it fails to cover as comprehensively as possible. Specifically, we design: (i) frequency-aware SfM that improves low-texture coverage via low-frequency view augmentation and relaxed multi-view correspondences; (ii) 3DGS self-initialization that lifts photometric supervision into additional points, compensating SfM-sparse regions with learned Gaussian centers; and (iii) point-cloud regularization that enforces multi-view consistency and uniform spatial coverage through simple geometric/visibility priors, yielding a clean and reliable point cloud. Our experiments on LLFF and Mip-NeRF360 demonstrate consistent gains in sparse-view settings, establishing our approach as a stronger initialization strategy. Code is available at https://github.com/zss171999645/ItG-GS.

Ren Zhang, Huilai Li, Chao qi et al.

Micro expression recognition (MER) is crucial for inferring genuine emotion. Applying a multimodal large language model (MLLM) to this task enables spatio-temporal analysis of facial motion and provides interpretable descriptions. However, there are still two core challenges: (1) The entanglement of static appearance and dynamic motion cues prevents the model from focusing on subtle motion; (2) Textual labels in existing MER datasets do not fully correspond to underlying facial muscle movements, creating a semantic gap between text supervision and physical motion. To address these issues, we propose DEFT-LLM, which achieves motion semantic alignment by multi-expert disentanglement. We first introduce Uni-MER, a motion-driven instruction dataset designed to align text with local facial motion. Its construction leverages dual constraints from optical flow and Action Unit (AU) labels to ensure spatio-temporal consistency and reasonable correspondence to the movements. We then design an architecture with three experts to decouple facial dynamics into independent and interpretable representations (structure, dynamic textures, and motion-semantics). By integrating the instruction-aligned knowledge from Uni-MER into DEFT-LLM, our method injects effective physical priors for micro expressions while also leveraging the cross modal reasoning ability of large language models, thus enabling precise capture of subtle emotional cues. Experiments on multiple challenging MER benchmarks demonstrate state-of-the-art performance, as well as a particular advantage in interpretable modeling of local facial motion.

Huilai Li, Yonghao Dang, Ying Xing et al.

Dense audio-visual event localization (DAVE) aims to identify event categories and locate the temporal boundaries in untrimmed videos. Most studies only employ event-related semantic constraints on the final outputs, lacking cross-modal semantic bridging in intermediate layers. This causes modality semantic gap for further fusion, making it difficult to distinguish between event-related content and irrelevant background content. Moreover, they rarely consider the correlations between events, which limits the model to infer concurrent events among complex scenarios. In this paper, we incorporate multi-stage semantic guidance and multi-event relationship modeling, which respectively enable hierarchical semantic understanding of audio-visual events and adaptive extraction of event dependencies, thereby better focusing on event-related information. Specifically, our eventaware semantic guided network (ESG-Net) includes a early semantics interaction (ESI) module and a mixture of dependency experts (MoDE) module. ESI applys multi-stage semantic guidance to explicitly constrain the model in learning semantic information through multi-modal early fusion and several classification loss functions, ensuring hierarchical understanding of event-related content. MoDE promotes the extraction of multi-event dependencies through multiple serial mixture of experts with adaptive weight allocation. Extensive experiments demonstrate that our method significantly surpasses the state-of-the-art methods, while greatly reducing parameters and computational load. Our code will be released on https://github.com/uchiha99999/ESG-Net.

Shaojie Zhang, Jianqin Yin, Yonghao Dang

Skeleton-based action recognition is a central task in computer vision and human-robot interaction. However, most previous methods suffer from overlooking the explicit exploitation of the latent data distributions (i.e., the intra-class variations and inter-class relations), thereby leading to confusion about ambiguous samples and sub-optimum solutions of the skeleton encoders. To mitigate this, we propose a Contrastive Spatiotemporal Representation Enhancement (CSRE) framework to obtain more discriminative representations from the sequences, which can be incorporated into various previous skeleton encoders and can be removed when testing. Specifically, we decompose the representation into spatial- and temporal-specific features to explore fine-grained motion patterns along the corresponding dimensions. Furthermore, to explicitly exploit the latent data distributions, we employ the attentive features to contrastive learning, which models the cross-sequence semantic relations by pulling together the features from the positive pairs and pushing away the negative pairs. Extensive experiments show that CSRE with five various skeleton encoders (HCN, 2S-AGCN, CTR-GCN, Hyperformer, and BlockGCN) achieves solid improvements on five benchmarks. The code will be released at https://github.com/zhshj0110/CSRE.

Xiaoli Liu, Jianqin Yin, Di Guo et al.

Early action prediction (EAP) aims to recognize human actions from a part of action execution in ongoing videos, which is an important task for many practical applications. Most prior works treat partial or full videos as a whole, ignoring rich action knowledge hidden in videos, i.e., semantic consistencies among different partial videos. In contrast, we partition original partial or full videos to form a new series of partial videos and mine the Action-Semantic Consistent Knowledge (ASCK) among these new partial videos evolving in arbitrary progress levels. Moreover, a novel Rich Action-semantic Consistent Knowledge network (RACK) under the teacher-student framework is proposed for EAP. Firstly, we use a two-stream pre-trained model to extract features of videos. Secondly, we treat the RGB or flow features of the partial videos as nodes and their action semantic consistencies as edges. Next, we build a bi-directional semantic graph for the teacher network and a single-directional semantic graph for the student network to model rich ASCK among partial videos. The MSE and MMD losses are incorporated as our distillation loss to enrich the ASCK of partial videos from the teacher to the student network. Finally, we obtain the final prediction by summering the logits of different subnetworks and applying a softmax layer. Extensive experiments and ablative studies have been conducted, demonstrating the effectiveness of modeling rich ASCK for EAP. With the proposed RACK, we have achieved state-of-the-art performance on three benchmarks. The code is available at https://github.com/lily2lab/RACK.git.

Yonghao Dang, Jianqin Yin, Shaojie Zhang

Video-based human pose estimation (VHPE) is a vital yet challenging task. While deep learning methods have made significant progress for the VHPE, most approaches to this task implicitly model the long-range interaction between joints by enlarging the receptive field of the convolution. Unlike prior methods, we design a lightweight and plug-and-play joint relation extractor (JRE) to model the associative relationship between joints explicitly and automatically. The JRE takes the pseudo heatmaps of joints as input and calculates the similarity between pseudo heatmaps. In this way, the JRE flexibly learns the relationship between any two joints, allowing it to learn the rich spatial configuration of human poses. Moreover, the JRE can infer invisible joints according to the relationship between joints, which is beneficial for the model to locate occluded joints. Then, combined with temporal semantic continuity modeling, we propose a Relation-based Pose Semantics Transfer Network (RPSTN) for video-based human pose estimation. Specifically, to capture the temporal dynamics of poses, the pose semantic information of the current frame is transferred to the next with a joint relation guided pose semantics propagator (JRPSP). The proposed model can transfer the pose semantic features from the non-occluded frame to the occluded frame, making our method robust to the occlusion. Furthermore, the proposed JRE module is also suitable for image-based human pose estimation. The proposed RPSTN achieves state-of-the-art results on the video-based Penn Action dataset, Sub-JHMDB dataset, and PoseTrack2018 dataset. Moreover, the proposed JRE improves the performance of backbones on the image-based COCO2017 dataset. Code is available at https://github.com/YHDang/pose-estimation.

Xiaoli Liu, Jianqin Yin, Huaping Liu et al.

Predicting future human motion plays a significant role in human-machine interactions for various real-life applications. A unified formulation and multi-order modeling are two critical perspectives for analyzing and representing human motion. In contrast to prior works, we improve the multi-order modeling ability of human motion systems for more accurate predictions by building a deep state-space model (DeepSSM). The DeepSSM utilizes the advantages of both the state-space theory and the deep network. Specifically, we formulate the human motion system as the state-space model of a dynamic system and model the motion system by the state-space theory, offering a unified formulation for diverse human motion systems. Moreover, a novel deep network is designed to parameterize this system, which jointly models the state-state transition and state-observation transition processes. In this way, the state of a system is updated by the multi-order information of a time-varying human motion sequence. Multiple future poses are recursively predicted via the state-observation transition. To further improve the model ability of the system, a novel loss, WT-MPJPE (Weighted Temporal Mean Per Joint Position Error), is introduced to optimize the model. The proposed loss encourages the system to achieve more accurate predictions by increasing weights to the early time steps. The experiments on two benchmark datasets (i.e., Human3.6M and 3DPW) confirm that our method achieves state-of-the-art performance with improved accuracy of at least 2.2mm per joint. The code will be available at: \url{https://github.com/lily2lab/DeepSSM.git}.

Huilai Li, Xiaomeng Di, Ying Xing et al.

Weakly supervised Audio-Visual Video Parsing (AVVP) aims to recognize and temporally localize audio, visual, and audio-visual events in videos using only coarse-grained labels. Faced with the challenging task settings, existing research advances along two main paths: pre-training pseudo-label generators for fine-grained cross-modal semantic guidance, or refining AVVP model architectures to enhance audio-visual fusion. However, since audio and visual signals are typically unaligned, achieving accurate video parsing fundamentally relies on precise perception of uni-modal events. Yet these multi-modal focused strategies excessively emphasize multi-modal fusion while inadequately guiding and preserving uni-modal semantics, resulting in noisy pseudo-labels and sub-optimal video parsing performance. This paper proposes a novel framework that enhances uni-modal representations for both the pseudo-label generator and the AVVP model. Specifically, we introduce a similarity-based label migration approach to annotate pre-training data, thereby enabling the pseudo-label generator to better understand uni-modal events. We also employ a soft-constrained manner to refine modeling of uni-modal features in parallel with multi-modal fusion. These designs enable coordinated attention to both uni-modal and cross-modal representations, thus boosting the localization performance for events. Extensive experiments show that our method outperforms state-of-the-art methods in both pseudo-label and AVVP performance.

Pengxiang Ding, Jianqin Yin

Joint relation modeling is a curial component in human motion prediction. Most existing methods rely on skeletal-based graphs to build the joint relations, where local interactive relations between joint pairs are well learned. However, the motion coordination, a global joint relation reflecting the simultaneous cooperation of all joints, is usually weakened because it is learned from part to whole progressively and asynchronously. Thus, the final predicted motions usually appear unrealistic. To tackle this issue, we learn a medium, called coordination attractor (CA), from the spatiotemporal features of motion to characterize the global motion features, which is subsequently used to build new relative joint relations. Through the CA, all joints are related simultaneously, and thus the motion coordination of all joints can be better learned. Based on this, we further propose a novel joint relation modeling module, Comprehensive Joint Relation Extractor (CJRE), to combine this motion coordination with the local interactions between joint pairs in a unified manner. Additionally, we also present a Multi-timescale Dynamics Extractor (MTDE) to extract enriched dynamics from the raw position information for effective prediction. Extensive experiments show that the proposed framework outperforms state-of-the-art methods in both short- and long-term predictions on H3.6M, CMU-Mocap, and 3DPW.

Feng Zhou, Jianqin Yin, Peiyang Li

The "lifting from 2D pose" method has been the dominant approach to 3D Human Pose Estimation (3DHPE) due to the powerful visual analysis ability of 2D pose estimators. Widely known, there exists a depth ambiguity problem when estimating solely from 2D pose, where one 2D pose can be mapped to multiple 3D poses. Intuitively, the rich semantic and texture information in images can contribute to a more accurate "lifting" procedure. Yet, existing research encounters two primary challenges. Firstly, the distribution of image data in 3D motion capture datasets is too narrow because of the laboratorial environment, which leads to poor generalization ability of methods trained with image information. Secondly, effective strategies for leveraging image information are lacking. In this paper, we give new insight into the cause of poor generalization problems and the effectiveness of image features. Based on that, we propose an advanced framework. Specifically, the framework consists of two stages. First, we enable the keypoints to query and select the beneficial features from all image patches. To reduce the keypoints attention to inconsequential background features, we design a novel Pose-guided Transformer Layer, which adaptively limits the updates to unimportant image patches. Then, through a designed Adaptive Feature Selection Module, we prune less significant image patches from the feature map. In the second stage, we allow the keypoints to further emphasize the retained critical image features. This progressive learning approach prevents further training on insignificant image features. Experimental results show that our model achieves state-of-the-art performance on both the Human3.6M dataset and the MPI-INF-3DHP dataset.

Shaojie Zhang, Jiahui Yang, Jianqin Yin et al.

Multimodal Large Language Models (MLLMs) have demonstrated significant success in visual understanding tasks. However, challenges persist in adapting these models for video comprehension due to the large volume of data and temporal complexity. Existing Video-LLMs using uniform frame sampling often struggle to capture the query-related crucial spatiotemporal clues of videos effectively. In this paper, we introduce Q-Frame, a novel approach for adaptive frame selection and multi-resolution scaling tailored to the video's content and the specific query. Q-Frame employs a training-free, plug-and-play strategy generated by a text-image matching network like CLIP, utilizing the Gumbel-Max trick for efficient frame selection. Q-Frame allows Video-LLMs to process more frames without exceeding computational limits, thereby preserving critical temporal and spatial information. We demonstrate Q-Frame's effectiveness through extensive experiments on benchmark datasets, including MLVU, LongVideoBench, and Video-MME, illustrating its superiority over existing methods and its applicability across various video understanding tasks.

Ruoqi Yin, Jianqin Yin

Human Interaction Recognition is the process of identifying interactive actions between multiple participants in a specific situation. The aim is to recognise the action interactions between multiple entities and their meaning. Many single Convolutional Neural Network has issues, such as the inability to capture global instance interaction features or difficulty in training, leading to ambiguity in action semantics. In addition, the computational complexity of the Transformer cannot be ignored, and its ability to capture local information and motion features in the image is poor. In this work, we propose a Two-stream Hybrid CNN-Transformer Network (THCT-Net), which exploits the local specificity of CNN and models global dependencies through the Transformer. CNN and Transformer simultaneously model the entity, time and space relationships between interactive entities respectively. Specifically, Transformer-based stream integrates 3D convolutions with multi-head self-attention to learn inter-token correlations; We propose a new multi-branch CNN framework for CNN-based streams that automatically learns joint spatio-temporal features from skeleton sequences. The convolutional layer independently learns the local features of each joint neighborhood and aggregates the features of all joints. And the raw skeleton coordinates as well as their temporal difference are integrated with a dual-branch paradigm to fuse the motion features of the skeleton. Besides, a residual structure is added to speed up training convergence. Finally, the recognition results of the two branches are fused using parallel splicing. Experimental results on diverse and challenging datasets, demonstrate that the proposed method can better comprehend and infer the meaning and context of various actions, outperforming state-of-the-art methods.

Yonghao Dang, Liyuan Liu, Hui Kang et al.

Real-time 2D keypoint detection plays an essential role in computer vision. Although CNN-based and Transformer-based methods have achieved breakthrough progress, they often fail to deliver superior performance and real-time speed. This paper introduces MamKPD, the first efficient yet effective mamba-based pose estimation framework for 2D keypoint detection. The conventional Mamba module exhibits limited information interaction between patches. To address this, we propose a lightweight contextual modeling module (CMM) that uses depth-wise convolutions to model inter-patch dependencies and linear layers to distill the pose cues within each patch. Subsequently, by combining Mamba for global modeling across all patches, MamKPD effectively extracts instances' pose information. We conduct extensive experiments on human and animal pose estimation datasets to validate the effectiveness of MamKPD. Our MamKPD-L achieves 77.3% AP on the COCO dataset with 1492 FPS on an NVIDIA GTX 4090 GPU. Moreover, MamKPD achieves state-of-the-art results on the MPII dataset and competitive results on the AP-10K dataset while saving 85% of the parameters compared to ViTPose. Our project page is available at https://mamkpd.github.io/.

Feng Zhou, Pu Cao, Yiyang Ma et al.

Denoising higher-resolution latents via a pre-trained U-Net leads to repetitive and disordered image patterns. Although recent studies make efforts to improve generative quality by aligning denoising process across original and higher resolutions, the root cause of suboptimal generation is still lacking exploration. Through comprehensive analysis of position encoding in U-Net, we attribute it to inconsistent position encoding, sourced by the inadequate propagation of position information from zero-padding to latent features in convolution layers as resolution increases. To address this issue, we propose a novel training-free approach, introducing a Progressive Boundary Complement (PBC) method. This method creates dynamic virtual image boundaries inside the feature map to enhance position information propagation, enabling high-quality and rich-content high-resolution image synthesis. Extensive experiments demonstrate the superiority of our method.

Zheng Zhou, Zhe Li, Bo Yu et al.

The automatic reconstruction of 3D computer-aided design (CAD) models from CAD sketches has recently gained significant attention in the computer vision community. Most existing methods, however, rely on vector CAD sketches and 3D ground truth for supervision, which are often difficult to be obtained in industrial applications and are sensitive to noise inputs. We propose viewing CAD reconstruction as a specific instance of sparse-view 3D reconstruction to overcome these limitations. While this reformulation offers a promising perspective, existing 3D reconstruction methods typically require natural images and corresponding camera poses as inputs, which introduces two major significant challenges: (1) modality discrepancy between CAD sketches and natural images, and (2) difficulty of accurate camera pose estimation for CAD sketches. To solve these issues, we first transform the CAD sketches into representations resembling natural images and extract corresponding masks. Next, we manually calculate the camera poses for the orthographic views to ensure accurate alignment within the 3D coordinate system. Finally, we employ a customized sparse-view 3D reconstruction method to achieve high-quality reconstructions from aligned orthographic views. By leveraging raster CAD sketches for self-supervision, our approach eliminates the reliance on vector CAD sketches and 3D ground truth. Experiments on the Sub-Fusion360 dataset demonstrate that our proposed method significantly outperforms previous approaches in CAD reconstruction performance and exhibits strong robustness to noisy inputs.

Yuanyuan Jiang, Jianqin Yin

While vision-language pretrained models (VLMs) excel in various multimodal understanding tasks, their potential in fine-grained audio-visual reasoning, particularly for audio-visual question answering (AVQA), remains largely unexplored. AVQA presents specific challenges for VLMs due to the requirement of visual understanding at the region level and seamless integration with audio modality. Previous VLM-based AVQA methods merely used CLIP as a feature encoder but underutilized its knowledge, and mistreated audio and video as separate entities in a dual-stream framework as most AVQA methods. This paper proposes a new CLIP-powered target-aware single-stream (TASS) network for AVQA using the image-text matching knowledge of the pretrained model through the audio-visual matching characteristic of nature. It consists of two key components: the target-aware spatial grounding module (TSG+) and the single-stream joint temporal grounding module (JTG). Specifically, we propose a TSG+ module to transfer the image-text matching knowledge from CLIP models to our region-text matching process without corresponding ground-truth labels. Moreover, unlike previous separate dual-stream networks that still required an additional audio-visual fusion module, JTG unifies audio-visual fusion and question-aware temporal grounding in a simplified single-stream architecture. It treats audio and video as a cohesive entity and further extends the pretrained image-text knowledge to audio-text matching by preserving their temporal correlation with our proposed cross-modal synchrony (CMS) loss. Extensive experiments conducted on the MUSIC-AVQA benchmark verified the effectiveness of our proposed method over existing state-of-the-art methods.

Yixun Zhang, Feng Zhou, Jianqin Yin

Camera-based perception is critical to autonomous driving yet remains vulnerable to task-specific adversarial manipulations in object detection and monocular depth estimation. Most existing 2D/3D attacks are developed in task silos, lack mechanisms to induce controllable depth bias, and offer no standardized protocol to quantify cross-task transfer, leaving the interaction between detection and depth underexplored. We present BiTAA, a bi-task adversarial attack built on 3D Gaussian Splatting that yields a single perturbation capable of simultaneously degrading detection and biasing monocular depth. Specifically, we introduce a dual-model attack framework that supports both full-image and patch settings and is compatible with common detectors and depth estimators, with optional expectation-over-transformation (EOT) for physical reality. In addition, we design a composite loss that couples detection suppression with a signed, magnitude-controlled log-depth bias within regions of interest (ROIs) enabling controllable near or far misperception while maintaining stable optimization across tasks. We also propose a unified evaluation protocol with cross-task transfer metrics and real-world evaluations, showing consistent cross-task degradation and a clear asymmetry between Det to Depth and from Depth to Det transfer. The results highlight practical risks for multi-task camera-only perception and motivate cross-task-aware defenses in autonomous driving scenarios.

Wei Wei, Shaojie Zhang, Yonghao Dang et al.

Human action recognition is a crucial task for intelligent robotics, particularly within the context of human-robot collaboration research. In self-supervised skeleton-based action recognition, the mask-based reconstruction paradigm learns the spatial structure and motion patterns of the skeleton by masking joints and reconstructing the target from unlabeled data. However, existing methods focus on a limited set of joints and low-order motion patterns, limiting the model's ability to understand complex motion patterns. To address this issue, we introduce MaskSem, a novel semantic-guided masking method for learning 3D hybrid high-order motion representations. This novel framework leverages Grad-CAM based on relative motion to guide the masking of joints, which can be represented as the most semantically rich temporal orgions. The semantic-guided masking process can encourage the model to explore more discriminative features. Furthermore, we propose using hybrid high-order motion as the reconstruction target, enabling the model to learn multi-order motion patterns. Specifically, low-order motion velocity and high-order motion acceleration are used together as the reconstruction target. This approach offers a more comprehensive description of the dynamic motion process, enhancing the model's understanding of motion patterns. Experiments on the NTU60, NTU120, and PKU-MMD datasets show that MaskSem, combined with a vanilla transformer, improves skeleton-based action recognition, making it more suitable for applications in human-robot interaction.

Yixun Zhang, Lizhi Wang, Junjun Zhao et al.

Camera-based object detection systems play a vital role in autonomous driving, yet they remain vulnerable to adversarial threats in real-world environments. Existing 2D and 3D physical attacks, due to their focus on texture optimization, often struggle to balance physical realism and attack robustness. In this work, we propose 3D Gaussian-based Adversarial Attack (3DGAA), a novel adversarial object generation framework that leverages the full 14-dimensional parameterization of 3D Gaussian Splatting (3DGS) to jointly optimize geometry and appearance in physically realizable ways. Unlike prior works that rely on patches or texture optimization, 3DGAA jointly perturbs both geometric attributes (shape, scale, rotation) and appearance attributes (color, opacity) to produce physically realistic and transferable adversarial objects. We further introduce a physical filtering module that filters outliers to preserve geometric fidelity, and a physical augmentation module that simulates complex physical scenarios to enhance attack generalization under real-world conditions. We evaluate 3DGAA on both virtual benchmarks and physical-world setups using miniature vehicle models. Experimental results show that 3DGAA achieves to reduce the detection mAP from 87.21\% to 7.38\%, significantly outperforming existing 3D physical attacks. Moreover, our method maintains high transferability across different physical conditions, demonstrating a new state-of-the-art in physically realizable adversarial attacks.

Chao Qi, Jianqin Yin, Ren Zhang

Image-point class incremental learning helps the 3D-points-vision robots continually learn category knowledge from 2D images, improving their perceptual capability in dynamic environments. However, some incremental learning methods address unimodal forgetting but fail in cross-modal cases, while others handle modal differences within training/testing datasets but assume no modal gaps between them. We first explore this cross-modal task, proposing a benchmark CMIP-CIL and relieving the cross-modal catastrophic forgetting problem. It employs masked point clouds and rendered multi-view images within a contrastive learning framework in pre-training, empowering the vision model with the generalizations of image-point correspondence. In the incremental stage, by freezing the backbone and promoting object representations close to their respective prototypes, the model effectively retains and generalizes knowledge across previously seen categories while continuing to learn new ones. We conduct comprehensive experiments on the benchmark datasets. Experiments prove that our method achieves state-of-the-art results, outperforming the baseline methods by a large margin.

Chao Qi, Jianqin Yin, Meng Chen et al.

Existing class-incremental learning methods in 3D point clouds rely on exemplars (samples of former classes) to resist the catastrophic forgetting of models, and exemplar-free settings will greatly degrade the performance. For exemplar-free incremental learning, the pre-trained model methods have achieved state-of-the-art results in 2D domains. However, these methods cannot be migrated to the 3D domains due to the limited pre-training datasets and insufficient focus on fine-grained geometric details. This paper breaks through these limitations, proposing a basic shape dataset with zero collection cost for model pre-training. It helps a model obtain extensive knowledge of 3D geometries. Based on this, we propose a framework embedded with 3D geometry knowledge for incremental learning in point clouds, compatible with exemplar-free (-based) settings. In the incremental stage, the geometry knowledge is extended to represent objects in point clouds. The class prototype is calculated by regularizing the data representation with the same category and is kept adjusting in the learning process. It helps the model remember the shape features of different categories. Experiments show that our method outperforms other baseline methods by a large margin on various benchmark datasets, considering both exemplar-free (-based) settings.

Guoliang Xu, Jianqin Yin, Ren Zhang et al.

Since COVID-19, crowd-counting tasks have gained wide applications. While supervised methods are reliable, annotation is more challenging in high-density scenes due to small head sizes and severe occlusion, whereas it's simpler in low-density scenes. Interestingly, can we train the model in low-density scenes and generalize it to high-density scenes? Therefore, we propose a low- to high-density generalization framework (L2HCount) that learns the pattern related to high-density scenes from low-density ones, enabling it to generalize well to high-density scenes. Specifically, we first introduce a High-Density Simulation Module and a Ground-Truth Generation Module to construct fake high-density images along with their corresponding ground-truth crowd annotations respectively by image-shifting technique, effectively simulating high-density crowd patterns. However, the simulated images have two issues: image blurring and loss of low-density image characteristics. Therefore, we second propose a Head Feature Enhancement Module to extract clear features in the simulated high-density scene. Third, we propose a Dual-Density Memory Encoding Module that uses two crowd memories to learn scene-specific patterns from low- and simulated high-density scenes, respectively. Extensive experiments on four challenging datasets have shown the promising performance of L2HCount.

Yuanyuan Jiang, Jianqin Yin

Audio-visual question answering (AVQA) is a challenging task that requires multistep spatio-temporal reasoning over multimodal contexts. Recent works rely on elaborate target-agnostic parsing of audio-visual scenes for spatial grounding while mistreating audio and video as separate entities for temporal grounding. This paper proposes a new target-aware joint spatio-temporal grounding network for AVQA. It consists of two key components: the target-aware spatial grounding module (TSG) and the single-stream joint audio-visual temporal grounding module (JTG). The TSG can focus on audio-visual cues relevant to the query subject by utilizing explicit semantics from the question. Unlike previous two-stream temporal grounding modules that required an additional audio-visual fusion module, JTG incorporates audio-visual fusion and question-aware temporal grounding into one module with a simpler single-stream architecture. The temporal synchronization between audio and video in the JTG is facilitated by our proposed cross-modal synchrony loss (CSL). Extensive experiments verified the effectiveness of our proposed method over existing state-of-the-art methods.

Chao Qi, Jianqin Yin

Uncertainty-aware semantic segmentation of the point clouds includes the predictive uncertainty estimation and the uncertainty-guided model optimization. One key challenge in the task is the efficiency of point-wise predictive distribution establishment. The widely-used MC dropout establishes the distribution by computing the standard deviation of samples using multiple stochastic forward propagations, which is time-consuming for tasks based on point clouds containing massive points. Hence, a framework embedded with NSA-MC dropout, a variant of MC dropout, is proposed to establish distributions in just one forward pass. Specifically, the NSA-MC dropout samples the model many times through a space-dependent way, outputting point-wise distribution by aggregating stochastic inference results of neighbors. Based on this, aleatoric and predictive uncertainties acquire from the predictive distribution. The aleatoric uncertainty is integrated into the loss function to penalize noisy points, avoiding the over-fitting of the model to some degree. Besides, the predictive uncertainty quantifies the confidence degree of predictions. Experimental results show that our framework obtains better segmentation results of real-world point clouds and efficiently quantifies the credibility of results. Our NSA-MC dropout is several times faster than MC dropout, and the inference time does not establish a coupling relation with the sampling times. The code will be available if the paper is accepted.

Mingshuai Dong, Shimin Wei, Jianqin Yin et al.

Recognizing the category of the object and using the features of the object itself to predict grasp configuration is of great significance to improve the accuracy of the grasp detection model and expand its application. Researchers have been trying to combine these capabilities in an end-to-end network to grasping specific objects in a cluttered scene efficiently. In this paper, we propose an end-to-end semantic grasp detection model, which can accomplish both semantic recognition and grasp detection. And we also design a target feature attention mechanism to guide the model focus on the features of target object ontology for grasp prediction according to the semantic information. This method effectively reduces the background features that are weakly correlated to the target object, thus making the features more unique and guaranteeing the accuracy and efficiency of grasp detection. Experimental results show that the proposed method can achieve 98.38% accuracy in Cornell Grasp Dataset. Furthermore, our results on complex multi-object scenarios or more rigorous evaluation metrics show the domain adaptability of our method over the state-of-the-art.

Xunli Zeng, Jianqin Yin

Amodal segmentation is a new direction of instance segmentation while considering the segmentation of the visible and occluded parts of the instance. The existing state-of-the-art method uses multi-task branches to predict the amodal part and the visible part separately and subtract the visible part from the amodal part to obtain the occluded part. However, the amodal part contains visible information. Therefore, the separated prediction method will generate duplicate information. Different from this method, we propose a method of amodal segmentation based on the idea of the jigsaw. The method uses multi-task branches to predict the two naturally decoupled parts of visible and occluded, which is like getting two matching jigsaw pieces. Then put the two jigsaw pieces together to get the amodal part. This makes each branch focus on the modeling of the object. And we believe that there are certain rules in the occlusion relationship in the real world. This is a kind of occlusion context information. This jigsaw method can better model the occlusion relationship and use the occlusion context information, which is important for amodal segmentation. Experiments on two widely used amodally annotated datasets prove that our method exceeds existing state-of-the-art methods. The source code of this work will be made public soon.

Pengxiang Ding, Jianqin Yin

Human motion prediction is essential for tasks such as human motion analysis and human-robot interactions. Most existing approaches have been proposed to realize motion prediction. However, they ignore an important task, the evaluation of the quality of the predicted result. It is far more enough for current approaches in actual scenarios because people can't know how to interact with the machine without the evaluation of prediction, and unreliable predictions may mislead the machine to harm the human. Hence, we propose an uncertainty-aware framework for human motion prediction (UA-HMP). Concretely, we first design an uncertainty-aware predictor through Gaussian modeling to achieve the value and the uncertainty of predicted motion. Then, an uncertainty-guided learning scheme is proposed to quantitate the uncertainty and reduce the negative effect of the noisy samples during optimization for better performance. Our proposed framework is easily combined with current SOTA baselines to overcome their weakness in uncertainty modeling with slight parameters increment. Extensive experiments also show that they can achieve better performance in both short and long-term predictions in H3.6M, CMU-Mocap.

Pengxiang Ding, Junying Wang, Jianqin Yin

Joint relation modeling is a curial component in human motion prediction. Most existing methods tend to design skeletal-based graphs to build the relations among joints, where local interactions between joint pairs are well learned. However, the global coordination of all joints, which reflects human motion's balance property, is usually weakened because it is learned from part to whole progressively and asynchronously. Thus, the final predicted motions are sometimes unnatural. To tackle this issue, we learn a medium, called balance attractor (BA), from the spatiotemporal features of motion to characterize the global motion features, which is subsequently used to build new joint relations. Through the BA, all joints are related synchronously, and thus the global coordination of all joints can be better learned. Based on the BA, we propose our framework, referred to Attractor-Guided Neural Network, mainly including Attractor-Based Joint Relation Extractor (AJRE) and Multi-timescale Dynamics Extractor (MTDE). The AJRE mainly includes Global Coordination Extractor (GCE) and Local Interaction Extractor (LIE). The former presents the global coordination of all joints, and the latter encodes local interactions between joint pairs. The MTDE is designed to extract dynamic information from raw position information for effective prediction. Extensive experiments show that the proposed framework outperforms state-of-the-art methods in both short and long-term predictions in H3.6M, CMU-Mocap, and 3DPW.