Yi Yang

Wenhao Wu, Xiaohan Wang, Haipeng Luo et al. · amazon-science

Vision-language models (VLMs) pre-trained on large-scale image-text pairs have demonstrated impressive transferability on various visual tasks. Transferring knowledge from such powerful VLMs is a promising direction for building effective video recognition models. However, current exploration in this field is still limited. We believe that the greatest value of pre-trained VLMs lies in building a bridge between visual and textual domains. In this paper, we propose a novel framework called BIKE, which utilizes the cross-modal bridge to explore bidirectional knowledge: i) We introduce the Video Attribute Association mechanism, which leverages the Video-to-Text knowledge to generate textual auxiliary attributes for complementing video recognition. ii) We also present a Temporal Concept Spotting mechanism that uses the Text-to-Video expertise to capture temporal saliency in a parameter-free manner, leading to enhanced video representation. Extensive studies on six popular video datasets, including Kinetics-400 & 600, UCF-101, HMDB-51, ActivityNet and Charades, show that our method achieves state-of-the-art performance in various recognition scenarios, such as general, zero-shot, and few-shot video recognition. Our best model achieves a state-of-the-art accuracy of 88.6% on the challenging Kinetics-400 using the released CLIP model. The code is available at https://github.com/whwu95/BIKE .

Jiancan Wu, Yi Yang, Yuchun Qian et al.

With the greater emphasis on privacy and security in our society, the problem of graph unlearning -- revoking the influence of specific data on the trained GNN model, is drawing increasing attention. However, ranging from machine unlearning to recently emerged graph unlearning methods, existing efforts either resort to retraining paradigm, or perform approximate erasure that fails to consider the inter-dependency between connected neighbors or imposes constraints on GNN structure, therefore hard to achieve satisfying performance-complexity trade-offs. In this work, we explore the influence function tailored for graph unlearning, so as to improve the unlearning efficacy and efficiency for graph unlearning. We first present a unified problem formulation of diverse graph unlearning tasks \wrt node, edge, and feature. Then, we recognize the crux to the inability of traditional influence function for graph unlearning, and devise Graph Influence Function (GIF), a model-agnostic unlearning method that can efficiently and accurately estimate parameter changes in response to a $ε$-mass perturbation in deleted data. The idea is to supplement the objective of the traditional influence function with an additional loss term of the influenced neighbors due to the structural dependency. Further deductions on the closed-form solution of parameter changes provide a better understanding of the unlearning mechanism. We conduct extensive experiments on four representative GNN models and three benchmark datasets to justify the superiority of GIF for diverse graph unlearning tasks in terms of unlearning efficacy, model utility, and unlearning efficiency. Our implementations are available at \url{https://github.com/wujcan/GIF-torch/}.

Chen Liang, Wenguan Wang, Tianfei Zhou et al.

We explore the task of language-guided video segmentation (LVS). Previous algorithms mostly adopt 3D CNNs to learn video representation, struggling to capture long-term context and easily suffering from visual-linguistic misalignment. In light of this, we present Locater (local-global context aware Transformer), which augments the Transformer architecture with a finite memory so as to query the entire video with the language expression in an efficient manner. The memory is designed to involve two components -- one for persistently preserving global video content, and one for dynamically gathering local temporal context and segmentation history. Based on the memorized local-global context and the particular content of each frame, Locater holistically and flexibly comprehends the expression as an adaptive query vector for each frame. The vector is used to query the corresponding frame for mask generation. The memory also allows Locater to process videos with linear time complexity and constant size memory, while Transformer-style self-attention computation scales quadratically with sequence length. To thoroughly examine the visual grounding capability of LVS models, we contribute a new LVS dataset, A2D-S+, which is built upon A2D-S dataset but poses increased challenges in disambiguating among similar objects. Experiments on three LVS datasets and our A2D-S+ show that Locater outperforms previous state-of-the-arts. Further, we won the 1st place in the Referring Video Object Segmentation Track of the 3rd Large-scale Video Object Segmentation Challenge, where Locater served as the foundation for the winning solution. Our code and dataset are available at: https://github.com/leonnnop/Locater

Kexin Li, Zongxin Yang, Lei Chen et al.

Audio-visual video segmentation~(AVVS) aims to generate pixel-level maps of sound-producing objects within image frames and ensure the maps faithfully adhere to the given audio, such as identifying and segmenting a singing person in a video. However, existing methods exhibit two limitations: 1) they address video temporal features and audio-visual interactive features separately, disregarding the inherent spatial-temporal dependence of combined audio and video, and 2) they inadequately introduce audio constraints and object-level information during the decoding stage, resulting in segmentation outcomes that fail to comply with audio directives. To tackle these issues, we propose a decoupled audio-video transformer that combines audio and video features from their respective temporal and spatial dimensions, capturing their combined dependence. To optimize memory consumption, we design a block, which, when stacked, enables capturing audio-visual fine-grained combinatorial-dependence in a memory-efficient manner. Additionally, we introduce audio-constrained queries during the decoding phase. These queries contain rich object-level information, ensuring the decoded mask adheres to the sounds. Experimental results confirm our approach's effectiveness, with our framework achieving a new SOTA performance on all three datasets using two backbones. The code is available at \url{https://github.com/aspirinone/CATR.github.io}

Xiaolong Shen, Zongxin Yang, Xiaohan Wang et al.

Video-based 3D human pose and shape estimations are evaluated by intra-frame accuracy and inter-frame smoothness. Although these two metrics are responsible for different ranges of temporal consistency, existing state-of-the-art methods treat them as a unified problem and use monotonous modeling structures (e.g., RNN or attention-based block) to design their networks. However, using a single kind of modeling structure is difficult to balance the learning of short-term and long-term temporal correlations, and may bias the network to one of them, leading to undesirable predictions like global location shift, temporal inconsistency, and insufficient local details. To solve these problems, we propose to structurally decouple the modeling of long-term and short-term correlations in an end-to-end framework, Global-to-Local Transformer (GLoT). First, a global transformer is introduced with a Masked Pose and Shape Estimation strategy for long-term modeling. The strategy stimulates the global transformer to learn more inter-frame correlations by randomly masking the features of several frames. Second, a local transformer is responsible for exploiting local details on the human mesh and interacting with the global transformer by leveraging cross-attention. Moreover, a Hierarchical Spatial Correlation Regressor is further introduced to refine intra-frame estimations by decoupled global-local representation and implicit kinematic constraints. Our GLoT surpasses previous state-of-the-art methods with the lowest model parameters on popular benchmarks, i.e., 3DPW, MPI-INF-3DHP, and Human3.6M. Codes are available at https://github.com/sxl142/GLoT.

Shuai Zhao, Linchao Zhu, Xiaohan Wang et al.

Recently, large-scale pre-training methods like CLIP have made great progress in multi-modal research such as text-video retrieval. In CLIP, transformers are vital for modeling complex multi-modal relations. However, in the vision transformer of CLIP, the essential visual tokenization process, which produces discrete visual token sequences, generates many homogeneous tokens due to the redundancy nature of consecutive and similar frames in videos. This significantly increases computation costs and hinders the deployment of video retrieval models in web applications. In this paper, to reduce the number of redundant video tokens, we design a multi-segment token clustering algorithm to find the most representative tokens and drop the non-essential ones. As the frame redundancy occurs mostly in consecutive frames, we divide videos into multiple segments and conduct segment-level clustering. Center tokens from each segment are later concatenated into a new sequence, while their original spatial-temporal relations are well maintained. We instantiate two clustering algorithms to efficiently find deterministic medoids and iteratively partition groups in high dimensional space. Through this token clustering and center selection procedure, we successfully reduce computation costs by removing redundant visual tokens. This method further enhances segment-level semantic alignment between video and text representations, enforcing the spatio-temporal interactions of tokens from within-segment frames. Our method, coined as CenterCLIP, surpasses existing state-of-the-art by a large margin on typical text-video benchmarks, while reducing the training memory cost by 35\% and accelerating the inference speed by 14\% at the best case. The code is available at \href{https://github.com/mzhaoshuai/CenterCLIP}{https://github.com/mzhaoshuai/CenterCLIP}.

Zongxin Yang, Jiaxu Miao, Yunchao Wei et al.

This paper delves into the challenges of achieving scalable and effective multi-object modeling for semi-supervised Video Object Segmentation (VOS). Previous VOS methods decode features with a single positive object, limiting the learning of multi-object representation as they must match and segment each target separately under multi-object scenarios. Additionally, earlier techniques catered to specific application objectives and lacked the flexibility to fulfill different speed-accuracy requirements. To address these problems, we present two innovative approaches, Associating Objects with Transformers (AOT) and Associating Objects with Scalable Transformers (AOST). In pursuing effective multi-object modeling, AOT introduces the IDentification (ID) mechanism to allocate each object a unique identity. This approach enables the network to model the associations among all objects simultaneously, thus facilitating the tracking and segmentation of objects in a single network pass. To address the challenge of inflexible deployment, AOST further integrates scalable long short-term transformers that incorporate scalable supervision and layer-wise ID-based attention. This enables online architecture scalability in VOS for the first time and overcomes ID embeddings' representation limitations. Given the absence of a benchmark for VOS involving densely multi-object annotations, we propose a challenging Video Object Segmentation in the Wild (VOSW) benchmark to validate our approaches. We evaluated various AOT and AOST variants using extensive experiments across VOSW and five commonly used VOS benchmarks, including YouTube-VOS 2018 & 2019 Val, DAVIS-2017 Val & Test, and DAVIS-2016. Our approaches surpass the state-of-the-art competitors and display exceptional efficiency and scalability consistently across all six benchmarks. Project page: https://github.com/yoxu515/aot-benchmark.

Changlin Li, Bohan Zhuang, Guangrun Wang et al.

Recent advances in vision Transformers (ViTs) have come with a voracious appetite for computing power, high-lighting the urgent need to develop efficient training methods for ViTs. Progressive learning, a training scheme where the model capacity grows progressively during training, has started showing its ability in efficient training. In this paper, we take a practical step towards efficient training of ViTs by customizing and automating progressive learning. First, we develop a strong manual baseline for progressive learning of ViTs, by introducing momentum growth (MoGrow) to bridge the gap brought by model growth. Then, we propose automated progressive learning (AutoProg), an efficient training scheme that aims to achieve lossless acceleration by automatically increasing the training overload on-the-fly; this is achieved by adaptively deciding whether, where and how much should the model grow during progressive learning. Specifically, we first relax the optimization of the growth schedule to sub-network architecture optimization problem, then propose one-shot estimation of the sub-network performance via an elastic supernet. The searching overhead is reduced to minimal by recycling the parameters of the supernet. Extensive experiments of efficient training on ImageNet with two representative ViT models, DeiT and VOLO, demonstrate that AutoProg can accelerate ViTs training by up to 85.1% with no performance drop. Code: https://github.com/changlin31/AutoProg

Fan Ma, Mike Zheng Shou, Linchao Zhu et al.

As an important area in computer vision, object tracking has formed two separate communities that respectively study Single Object Tracking (SOT) and Multiple Object Tracking (MOT). However, current methods in one tracking scenario are not easily adapted to the other due to the divergent training datasets and tracking objects of both tasks. Although UniTrack \cite{wang2021different} demonstrates that a shared appearance model with multiple heads can be used to tackle individual tracking tasks, it fails to exploit the large-scale tracking datasets for training and performs poorly on single object tracking. In this work, we present the Unified Transformer Tracker (UTT) to address tracking problems in different scenarios with one paradigm. A track transformer is developed in our UTT to track the target in both SOT and MOT. The correlation between the target and tracking frame features is exploited to localize the target. We demonstrate that both SOT and MOT tasks can be solved within this framework. The model can be simultaneously end-to-end trained by alternatively optimizing the SOT and MOT objectives on the datasets of individual tasks. Extensive experiments are conducted on several benchmarks with a unified model trained on SOT and MOT datasets. Code will be available at https://github.com/Flowerfan/Trackron.

Wenhao Wang, Yifan Sun, Wei Li et al.

This paper explores a hierarchical prompting mechanism for the hierarchical image classification (HIC) task. Different from prior HIC methods, our hierarchical prompting is the first to explicitly inject ancestor-class information as a tokenized hint that benefits the descendant-class discrimination. We think it well imitates human visual recognition, i.e., humans may use the ancestor class as a prompt to draw focus on the subtle differences among descendant classes. We model this prompting mechanism into a Transformer with Hierarchical Prompting (TransHP). TransHP consists of three steps: 1) learning a set of prompt tokens to represent the coarse (ancestor) classes, 2) on-the-fly predicting the coarse class of the input image at an intermediate block, and 3) injecting the prompt token of the predicted coarse class into the intermediate feature. Though the parameters of TransHP maintain the same for all input images, the injected coarse-class prompt conditions (modifies) the subsequent feature extraction and encourages a dynamic focus on relatively subtle differences among the descendant classes. Extensive experiments show that TransHP improves image classification on accuracy (e.g., improving ViT-B/16 by +2.83% ImageNet classification accuracy), training data efficiency (e.g., +12.69% improvement under 10% ImageNet training data), and model explainability. Moreover, TransHP also performs favorably against prior HIC methods, showing that TransHP well exploits the hierarchical information. The code is available at: https://github.com/WangWenhao0716/TransHP.

Mel Vecerik, Carl Doersch, Yi Yang et al. · deepmind

For robots to be useful outside labs and specialized factories we need a way to teach them new useful behaviors quickly. Current approaches lack either the generality to onboard new tasks without task-specific engineering, or else lack the data-efficiency to do so in an amount of time that enables practical use. In this work we explore dense tracking as a representational vehicle to allow faster and more general learning from demonstration. Our approach utilizes Track-Any-Point (TAP) models to isolate the relevant motion in a demonstration, and parameterize a low-level controller to reproduce this motion across changes in the scene configuration. We show this results in robust robot policies that can solve complex object-arrangement tasks such as shape-matching, stacking, and even full path-following tasks such as applying glue and sticking objects together, all from demonstrations that can be collected in minutes.

Wei Shang, Dongwei Ren, Yi Yang et al.

Video deblurring methods, aiming at recovering consecutive sharp frames from a given blurry video, usually assume that the input video suffers from consecutively blurry frames. However, in real-world scenarios captured by modern imaging devices, sharp frames often interspersed within the video, providing temporally nearest sharp features that can aid in the restoration of blurry frames. In this work, we propose a video deblurring method that leverages both neighboring frames and existing sharp frames using hybrid Transformers for feature aggregation. Specifically, we first train a blur-aware detector to distinguish between sharp and blurry frames. Then, a window-based local Transformer is employed for exploiting features from neighboring frames, where cross attention is beneficial for aggregating features from neighboring frames without explicit spatial alignment. To aggregate nearest sharp features from detected sharp frames, we utilize a global Transformer with multi-scale matching capability. Moreover, our method can easily be extended to event-driven video deblurring by incorporating an event fusion module into the global Transformer. Extensive experiments on benchmark datasets demonstrate that our proposed method outperforms state-of-the-art video deblurring methods as well as event-driven video deblurring methods in terms of quantitative metrics and visual quality. The source code and trained models are available at https://github.com/shangwei5/STGTN.

Qingwen Zhang, Yi Yang, Peizheng Li et al.

Scene flow estimation predicts the 3D motion at each point in successive LiDAR scans. This detailed, point-level, information can help autonomous vehicles to accurately predict and understand dynamic changes in their surroundings. Current state-of-the-art methods require annotated data to train scene flow networks and the expense of labeling inherently limits their scalability. Self-supervised approaches can overcome the above limitations, yet face two principal challenges that hinder optimal performance: point distribution imbalance and disregard for object-level motion constraints. In this paper, we propose SeFlow, a self-supervised method that integrates efficient dynamic classification into a learning-based scene flow pipeline. We demonstrate that classifying static and dynamic points helps design targeted objective functions for different motion patterns. We also emphasize the importance of internal cluster consistency and correct object point association to refine the scene flow estimation, in particular on object details. Our real-time capable method achieves state-of-the-art performance on the self-supervised scene flow task on Argoverse 2 and Waymo datasets. The code is open-sourced at https://github.com/KTH-RPL/SeFlow along with trained model weights.

Xiaofei Wang, Mingliang Han, Tianyu Hao et al.

Vision-language-action (VLA) models are gaining attention in robotics, yet their robustness to adversarial attacks remains largely unexplored. Existing work shows that adversarial patches can mislead VLA-based robots but assumes full access to the entire execution trajectory, an unrealistic requirement in practice. We address this limitation by formulating a partially observable threat model, where the adversary can exploit only a short prefix of the trajectory to generate a fixed patch applied to all subsequent frames. Under this setting, we propose a two-phase framework. First, we localize the patch using the model's attention maps to identify visually critical regions that correspond to the full instruction. Then, we optimize the patch to disrupt the semantic grounding of target objects and increase the curvature of action trajectories, thereby compounding failures in both perception and control. Extensive experiments in simulation and real-world robotic environments show that our method sustains adversarial effects under partial observability, inducing long-horizon disruptions and significantly reducing task success rates.

Zongxin Yang, Yi Yang

This paper focuses on developing a more effective method of hierarchical propagation for semi-supervised Video Object Segmentation (VOS). Based on vision transformers, the recently-developed Associating Objects with Transformers (AOT) approach introduces hierarchical propagation into VOS and has shown promising results. The hierarchical propagation can gradually propagate information from past frames to the current frame and transfer the current frame feature from object-agnostic to object-specific. However, the increase of object-specific information will inevitably lead to the loss of object-agnostic visual information in deep propagation layers. To solve such a problem and further facilitate the learning of visual embeddings, this paper proposes a Decoupling Features in Hierarchical Propagation (DeAOT) approach. Firstly, DeAOT decouples the hierarchical propagation of object-agnostic and object-specific embeddings by handling them in two independent branches. Secondly, to compensate for the additional computation from dual-branch propagation, we propose an efficient module for constructing hierarchical propagation, i.e., Gated Propagation Module, which is carefully designed with single-head attention. Extensive experiments show that DeAOT significantly outperforms AOT in both accuracy and efficiency. On YouTube-VOS, DeAOT can achieve 86.0% at 22.4fps and 82.0% at 53.4fps. Without test-time augmentations, we achieve new state-of-the-art performance on four benchmarks, i.e., YouTube-VOS (86.2%), DAVIS 2017 (86.2%), DAVIS 2016 (92.9%), and VOT 2020 (0.622). Project page: https://github.com/z-x-yang/AOT.

Yuxuan Hou, Hehe Fan, Jianrong Zhang et al.

The automatic synthesis of analog circuits presents significant challenges. Most existing approaches formulate the problem as a single-objective optimization task, overlooking that design specifications for a given circuit type vary widely across applications. To address this, we introduce specification-conditioned analog circuit generation, a task that directly generates analog circuits based on target specifications. The motivation is to leverage existing well-designed circuits to improve automation in analog circuit design. Specifically, we propose CktGen, a simple yet effective variational autoencoder that maps discretized specifications and circuits into a joint latent space and reconstructs the circuit from that latent vector. Notably, as a single specification may correspond to multiple valid circuits, naively fusing specification information into the generative model does not capture these one-to-many relationships. To address this, we decouple the encoding of circuits and specifications and align their mapped latent space. Then, we employ contrastive training with a filter mask to maximize differences between encoded circuits and specifications. Furthermore, classifier guidance along with latent feature alignment promotes the clustering of circuits sharing the same specification, avoiding model collapse into trivial one-to-one mappings. By canonicalizing the latent space with respect to specifications, we can search for an optimal circuit that meets valid target specifications. We conduct comprehensive experiments on the open circuit benchmark and introduce metrics to evaluate cross-model consistency. Experimental results demonstrate that CktGen achieves substantial improvements over state-of-the-art methods.

Zechuan Zhang, Li Sun, Zongxin Yang et al.

Reconstructing 3D clothed human avatars from single images is a challenging task, especially when encountering complex poses and loose clothing. Current methods exhibit limitations in performance, largely attributable to their dependence on insufficient 2D image features and inconsistent query methods. Owing to this, we present the Global-correlated 3D-decoupling Transformer for clothed Avatar reconstruction (GTA), a novel transformer-based architecture that reconstructs clothed human avatars from monocular images. Our approach leverages transformer architectures by utilizing a Vision Transformer model as an encoder for capturing global-correlated image features. Subsequently, our innovative 3D-decoupling decoder employs cross-attention to decouple tri-plane features, using learnable embeddings as queries for cross-plane generation. To effectively enhance feature fusion with the tri-plane 3D feature and human body prior, we propose a hybrid prior fusion strategy combining spatial and prior-enhanced queries, leveraging the benefits of spatial localization and human body prior knowledge. Comprehensive experiments on CAPE and THuman2.0 datasets illustrate that our method outperforms state-of-the-art approaches in both geometry and texture reconstruction, exhibiting high robustness to challenging poses and loose clothing, and producing higher-resolution textures. Codes will be available at https://github.com/River-Zhang/GTA.

Feng Zhu, Zongxin Yang, Xin Yu et al.

Modeling temporal information for both detection and tracking in a unified framework has been proved a promising solution to video instance segmentation (VIS). However, how to effectively incorporate the temporal information into an online model remains an open problem. In this work, we propose a new online VIS paradigm named Instance As Identity (IAI), which models temporal information for both detection and tracking in an efficient way. In detail, IAI employs a novel identification module to predict identification number for tracking instances explicitly. For passing temporal information cross frame, IAI utilizes an association module which combines current features and past embeddings. Notably, IAI can be integrated with different image models. We conduct extensive experiments on three VIS benchmarks. IAI outperforms all the online competitors on YouTube-VIS-2019 (ResNet-101 43.7 mAP) and YouTube-VIS-2021 (ResNet-50 38.0 mAP). Surprisingly, on the more challenging OVIS, IAI achieves SOTA performance (20.6 mAP). Code is available at https://github.com/zfonemore/IAI

Xiao Pan, Peike Li, Zongxin Yang et al.

In this paper, we focus on unsupervised learning for Video Object Segmentation (VOS) which learns visual correspondence (i.e., the similarity between pixel-level features) from unlabeled videos. Previous methods are mainly based on the contrastive learning paradigm, which optimize either in image level or pixel level. Image-level optimization (e.g., the spatially pooled feature of ResNet) learns robust high-level semantics but is sub-optimal since the pixel-level features are optimized implicitly. By contrast, pixel-level optimization is more explicit, however, it is sensitive to the visual quality of training data and is not robust to object deformation. To complementarily perform these two levels of optimization in a unified framework, we propose the In-aNd-Out (INO) generative learning from a purely generative perspective with the help of naturally designed class tokens and patch tokens in Vision Transformer (ViT). Specifically, for image-level optimization, we force the out-view imagination from local to global views on class tokens, which helps capture high-level semantics, and we name it as out-generative learning. As to pixel-level optimization, we perform in-view masked image modeling on patch tokens, which recovers the corrupted parts of an image via inferring its fine-grained structure, and we term it as in-generative learning. To discover the temporal information better, we additionally force the inter-frame consistency from both feature and affinity matrix levels. Extensive experiments on DAVIS-2017 val and YouTube-VOS 2018 val show that our INO outperforms previous state-of-the-art methods by significant margins. Code is available: https://github.com/pansanity666/INO_VOS

Chishui Chen, Jiaye Lin, Te Sun et al.

Agent skills are callable procedural modules that provide reusable knowledge and execution policies for complex agentic tasks. However, existing methods mainly focus on selecting relevant skills or improving the skills themselves, while overlooking whether a relevant skill should actually be invoked at the current decision point. Unhelpful invocations may introduce irrelevant context and disrupt an otherwise correct execution process. To address this issue, we propose SelSkill, a dual-granularity preference-learning framework for selective skill invocation. SelSkill formulates skill use as a skill-or-skip decision, uses predictive uncertainty to prioritize candidate decision points, and constructs controlled invoke-skip preference pairs from shared trajectory prefixes. It further combines episode-level outcome preferences with step-level invocation preferences to capture both overall trajectory quality and the local effectiveness of skill invocation. On ALFWorld with Qwen3-8B, SelSkill improves task success by 10.9 percentage points and execution precision by 29.1 percentage points. On BFCL, it improves task success by 5.7 percentage points and execution precision by 29.5 percentage points. Zero-shot results on Tau-bench and PopQA further suggest that the learned invocation policy transfers to new domains with previously unseen skills.

Carl Doersch, Yi Yang, Mel Vecerik et al.

We present a novel model for Tracking Any Point (TAP) that effectively tracks any queried point on any physical surface throughout a video sequence. Our approach employs two stages: (1) a matching stage, which independently locates a suitable candidate point match for the query point on every other frame, and (2) a refinement stage, which updates both the trajectory and query features based on local correlations. The resulting model surpasses all baseline methods by a significant margin on the TAP-Vid benchmark, as demonstrated by an approximate 20% absolute average Jaccard (AJ) improvement on DAVIS. Our model facilitates fast inference on long and high-resolution video sequences. On a modern GPU, our implementation has the capacity to track points faster than real-time, and can be flexibly extended to higher-resolution videos. Given the high-quality trajectories extracted from a large dataset, we demonstrate a proof-of-concept diffusion model which generates trajectories from static images, enabling plausible animations. Visualizations, source code, and pretrained models can be found on our project webpage.

Carl Doersch, Ankush Gupta, Larisa Markeeva et al.

Generic motion understanding from video involves not only tracking objects, but also perceiving how their surfaces deform and move. This information is useful to make inferences about 3D shape, physical properties and object interactions. While the problem of tracking arbitrary physical points on surfaces over longer video clips has received some attention, no dataset or benchmark for evaluation existed, until now. In this paper, we first formalize the problem, naming it tracking any point (TAP). We introduce a companion benchmark, TAP-Vid, which is composed of both real-world videos with accurate human annotations of point tracks, and synthetic videos with perfect ground-truth point tracks. Central to the construction of our benchmark is a novel semi-automatic crowdsourced pipeline which uses optical flow estimates to compensate for easier, short-term motion like camera shake, allowing annotators to focus on harder sections of video. We validate our pipeline on synthetic data and propose a simple end-to-end point tracking model TAP-Net, showing that it outperforms all prior methods on our benchmark when trained on synthetic data.

Zhedong Zheng, Jiayin Zhu, Wei Ji et al.

This research aims to study a self-supervised 3D clothing reconstruction method, which recovers the geometry shape and texture of human clothing from a single image. Compared with existing methods, we observe that three primary challenges remain: (1) 3D ground-truth meshes of clothing are usually inaccessible due to annotation difficulties and time costs; (2) Conventional template-based methods are limited to modeling non-rigid objects, e.g., handbags and dresses, which are common in fashion images; (3) The inherent ambiguity compromises the model training, such as the dilemma between a large shape with a remote camera or a small shape with a close camera. In an attempt to address the above limitations, we propose a causality-aware self-supervised learning method to adaptively reconstruct 3D non-rigid objects from 2D images without 3D annotations. In particular, to solve the inherent ambiguity among four implicit variables, i.e., camera position, shape, texture, and illumination, we introduce an explainable structural causal map (SCM) to build our model. The proposed model structure follows the spirit of the causal map, which explicitly considers the prior template in the camera estimation and shape prediction. When optimization, the causality intervention tool, i.e., two expectation-maximization loops, is deeply embedded in our algorithm to (1) disentangle four encoders and (2) facilitate the prior template. Extensive experiments on two 2D fashion benchmarks (ATR and Market-HQ) show that the proposed method could yield high-fidelity 3D reconstruction. Furthermore, we also verify the scalability of the proposed method on a fine-grained bird dataset, i.e., CUB. The code is available at https://github.com/layumi/ 3D-Magic-Mirror .

Shuangkang Fang, Weixin Xu, Heng Wang et al.

Neural Radiance Fields (NeRF) methods have proved effective as compact, high-quality and versatile representations for 3D scenes, and enable downstream tasks such as editing, retrieval, navigation, etc. Various neural architectures are vying for the core structure of NeRF, including the plain Multi-Layer Perceptron (MLP), sparse tensors, low-rank tensors, hashtables and their compositions. Each of these representations has its particular set of trade-offs. For example, the hashtable-based representations admit faster training and rendering but their lack of clear geometric meaning hampers downstream tasks like spatial-relation-aware editing. In this paper, we propose Progressive Volume Distillation (PVD), a systematic distillation method that allows any-to-any conversions between different architectures, including MLP, sparse or low-rank tensors, hashtables and their compositions. PVD consequently empowers downstream applications to optimally adapt the neural representations for the task at hand in a post hoc fashion. The conversions are fast, as distillation is progressively performed on different levels of volume representations, from shallower to deeper. We also employ special treatment of density to deal with its specific numerical instability problem. Empirical evidence is presented to validate our method on the NeRF-Synthetic, LLFF and TanksAndTemples datasets. For example, with PVD, an MLP-based NeRF model can be distilled from a hashtable-based Instant-NGP model at a 10X~20X faster speed than being trained the original NeRF from scratch, while achieving a superior level of synthesis quality. Code is available at https://github.com/megvii-research/AAAI2023-PVD.

Shuai Zhao, Linchao Zhu, Xiaohan Wang et al.

Self-supervised learning makes significant progress in pre-training large models, but struggles with small models. Mainstream solutions to this problem rely mainly on knowledge distillation, which involves a two-stage procedure: first training a large teacher model and then distilling it to improve the generalization ability of smaller ones. In this work, we introduce another one-stage solution to obtain pre-trained small models without the need for extra teachers, namely, slimmable networks for contrastive self-supervised learning (SlimCLR). A slimmable network consists of a full network and several weight-sharing sub-networks, which can be pre-trained once to obtain various networks, including small ones with low computation costs. However, interference between weight-sharing networks leads to severe performance degradation in self-supervised cases, as evidenced by gradient magnitude imbalance and gradient direction divergence. The former indicates that a small proportion of parameters produce dominant gradients during backpropagation, while the main parameters may not be fully optimized. The latter shows that the gradient direction is disordered, and the optimization process is unstable. To address these issues, we introduce three techniques to make the main parameters produce dominant gradients and sub-networks have consistent outputs. These techniques include slow start training of sub-networks, online distillation, and loss re-weighting according to model sizes. Furthermore, theoretical results are presented to demonstrate that a single slimmable linear layer is sub-optimal during linear evaluation. Thus a switchable linear probe layer is applied during linear evaluation. We instantiate SlimCLR with typical contrastive learning frameworks and achieve better performance than previous arts with fewer parameters and FLOPs. The code is at https://github.com/mzhaoshuai/SlimCLR.

Yuanyou Xu, Zongxin Yang, Yi Yang

Tracking any given object(s) spatially and temporally is a common purpose in Visual Object Tracking (VOT) and Video Object Segmentation (VOS). Joint tracking and segmentation have been attempted in some studies but they often lack full compatibility of both box and mask in initialization and prediction, and mainly focus on single-object scenarios. To address these limitations, this paper proposes a Multi-object Mask-box Integrated framework for unified Tracking and Segmentation, dubbed MITS. Firstly, the unified identification module is proposed to support both box and mask reference for initialization, where detailed object information is inferred from boxes or directly retained from masks. Additionally, a novel pinpoint box predictor is proposed for accurate multi-object box prediction, facilitating target-oriented representation learning. All target objects are processed simultaneously from encoding to propagation and decoding, as a unified pipeline for VOT and VOS. Experimental results show MITS achieves state-of-the-art performance on both VOT and VOS benchmarks. Notably, MITS surpasses the best prior VOT competitor by around 6% on the GOT-10k test set, and significantly improves the performance of box initialization on VOS benchmarks. The code is available at https://github.com/yoxu515/MITS.

Jing Peng, Junhao Du, Chenghao Wang et al.

Speech foundation models and Speech LLMs have advanced speech understanding, yet deployment-oriented model selection is hindered by non-comparable evaluations caused by mismatched post-processing, and by training results that are hard to reproduce across data scales and pipelines. We present SURE, a unified experimentation framework that standardizes prediction formats, normalization, and scoring. SURE evaluates strong systems across paradigms, from conventional pipelines to Speech LLMs, on representative tasks under realistic acoustic and linguistic stressors. Beyond evaluation, SURE introduces an agent-assisted training conversion flow that maps paper and code into versioned, runnable training pipelines under a unified protocol on matched open-data subsets. Overall, SURE improves comparability and reproducibility for deployment-oriented evaluation.

Yixuan Tang, Yi Yang

AI research agents can now generate research ideas, design experiments, run code, and draft papers, raising the possibility of large-scale AI-assisted scientific discovery. Many current agent frameworks explicitly encourage the generation of novel and high-impact ideas. Yet it remains unclear whether AI-assisted ideation broadens scientific exploration or mainly concentrates around existing work. We study AI research agents as scientific search systems. Using four AI research-agent frameworks and six large language models, we generate 37,802 scientific ideas from shared seed literature across citation-defined research areas in AI and machine learning. We then compare the resulting AI ideas against human-authored papers from the same research areas, follow-on human research emerging from the same seed literature, and the seed literature itself. Across experiments, four consistent patterns emerge. First, AI-generated ideas are substantially more concentrated than human-authored papers from the same research areas. Second, AI-generated ideas remain much closer to their starting literature than later human follow-on work does. Third, papers most similar to AI-generated ideas tend to receive lower subsequent citations. Fourth, when AI-generated ideas differ from prior work, the differences arise primarily from recombining existing technical methods rather than introducing fundamentally new research questions. Overall, current AI research agents appear better suited to local elaboration than to broadening scientific exploration.

Kimi Team, Tongtong Bai, Yifan Bai et al.

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to $4.5\times$ over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence.

Shuangkang Fang, Yufeng Wang, Yi-Hsuan Tsai et al.

Recent work on image content manipulation based on vision-language pre-training models has been effectively extended to text-driven 3D scene editing. However, existing schemes for 3D scene editing still exhibit certain shortcomings, hindering their further interactive design. Such schemes typically adhere to fixed input patterns, limiting users' flexibility in text input. Moreover, their editing capabilities are constrained by a single or a few 2D visual models and require intricate pipeline design to integrate these models into 3D reconstruction processes. To address the aforementioned issues, we propose a dialogue-based 3D scene editing approach, termed CE3D, which is centered around a large language model that allows for arbitrary textual input from users and interprets their intentions, subsequently facilitating the autonomous invocation of the corresponding visual expert models. Furthermore, we design a scheme utilizing Hash-Atlas to represent 3D scene views, which transfers the editing of 3D scenes onto 2D atlas images. This design achieves complete decoupling between the 2D editing and 3D reconstruction processes, enabling CE3D to flexibly integrate a wide range of existing 2D or 3D visual models without necessitating intricate fusion designs. Experimental results demonstrate that CE3D effectively integrates multiple visual models to achieve diverse editing visual effects, possessing strong scene comprehension and multi-round dialog capabilities. The code is available at https://sk-fun.fun/CE3D.

Yi Yang, Zhihong Liu, Siqi Kou et al.

We propose world-language-action (WLA) models as a new class of embodied foundation models. WLA takes textual instructions, images, and robot states as inputs to jointly predict textual subtasks, subgoal images, and robot actions, conjoining the \emph{world modeling interface} to learn from extensive egocentric videos as in the world-action model (WAM) and the \emph{language reasoning} capacities to solve complex long-horizon tasks as in vision-language-action (VLA) models. At the core of WLA lies an \emph{autoregressive (AR)} Transformer backbone, instead of a bidirectional diffusion Transformer as in WAMs, to predict the \emph{next state}, comprising the \emph{semantic-level} textual intention and complementary \emph{fine-grained} physical dynamics. The physical dynamics are supervised by the world modeling objective based on a dedicated World Expert, and are leveraged to ease the characterization of the state-action correlation for the Action Expert. WLA leverages meta-queries to make the world prediction \emph{implicitly} impact the action generation so that the former can be disabled during inference. The world prediction can also be activated to enable test-time scaling for improved robot control. Our WLA-0 prototype, with 2B active parameters, achieves 40 ms per inference on an NVIDIA RTX 5090. Evaluations across simulated and real-world environments demonstrate that WLA-0 achieves state-of-the-art multi-task and long-horizon learning abilities, e.g., 92.94\% success rate on RoboTwin2.0 Clean and 56.5\% success rate on RMBench. WLA-0 also holds the promise to learn novel tasks directly from \emph{cross-embodiment robot videos} without action annotations.

Dewei Zhou, You Li, Fan Ma et al.

We introduce the Multi-Instance Generation (MIG) task, which focuses on generating multiple instances within a single image, each accurately placed at predefined positions with attributes such as category, color, and shape, strictly following user specifications. MIG faces three main challenges: avoiding attribute leakage between instances, supporting diverse instance descriptions, and maintaining consistency in iterative generation. To address attribute leakage, we propose the Multi-Instance Generation Controller (MIGC). MIGC generates multiple instances through a divide-and-conquer strategy, breaking down multi-instance shading into single-instance tasks with singular attributes, later integrated. To provide more types of instance descriptions, we developed MIGC++. MIGC++ allows attribute control through text \& images and position control through boxes \& masks. Lastly, we introduced the Consistent-MIG algorithm to enhance the iterative MIG ability of MIGC and MIGC++. This algorithm ensures consistency in unmodified regions during the addition, deletion, or modification of instances, and preserves the identity of instances when their attributes are changed. We introduce the COCO-MIG and Multimodal-MIG benchmarks to evaluate these methods. Extensive experiments on these benchmarks, along with the COCO-Position benchmark and DrawBench, demonstrate that our methods substantially outperform existing techniques, maintaining precise control over aspects including position, attribute, and quantity. Project page: https://github.com/limuloo/MIGC.

Liulei Li, Tianfei Zhou, Wenguan Wang et al.

Humans are able to recognize structured relations in observation, allowing us to decompose complex scenes into simpler parts and abstract the visual world in multiple levels. However, such hierarchical reasoning ability of human perception remains largely unexplored in current literature of semantic segmentation. Existing work is often aware of flatten labels and predicts target classes exclusively for each pixel. In this paper, we instead address hierarchical semantic segmentation (HSS), which aims at structured, pixel-wise description of visual observation in terms of a class hierarchy. We devise HSSN, a general HSS framework that tackles two critical issues in this task: i) how to efficiently adapt existing hierarchy-agnostic segmentation networks to the HSS setting, and ii) how to leverage the hierarchy information to regularize HSS network learning. To address i), HSSN directly casts HSS as a pixel-wise multi-label classification task, only bringing minimal architecture change to current segmentation models. To solve ii), HSSN first explores inherent properties of the hierarchy as a training objective, which enforces segmentation predictions to obey the hierarchy structure. Further, with hierarchy-induced margin constraints, HSSN reshapes the pixel embedding space, so as to generate well-structured pixel representations and improve segmentation eventually. We conduct experiments on four semantic segmentation datasets (i.e., Mapillary Vistas 2.0, Cityscapes, LIP, and PASCAL-Person-Part), with different class hierarchies, segmentation network architectures and backbones, showing the generalization and superiority of HSSN.

Chen Liang, Wenguan Wang, Jiaxu Miao et al.

Prevalent semantic segmentation solutions are, in essence, a dense discriminative classifier of p(class|pixel feature). Though straightforward, this de facto paradigm neglects the underlying data distribution p(pixel feature|class), and struggles to identify out-of-distribution data. Going beyond this, we propose GMMSeg, a new family of segmentation models that rely on a dense generative classifier for the joint distribution p(pixel feature,class). For each class, GMMSeg builds Gaussian Mixture Models (GMMs) via Expectation-Maximization (EM), so as to capture class-conditional densities. Meanwhile, the deep dense representation is end-to-end trained in a discriminative manner, i.e., maximizing p(class|pixel feature). This endows GMMSeg with the strengths of both generative and discriminative models. With a variety of segmentation architectures and backbones, GMMSeg outperforms the discriminative counterparts on three closed-set datasets. More impressively, without any modification, GMMSeg even performs well on open-world datasets. We believe this work brings fundamental insights into the related fields.

Tingyu Wang, Zhedong Zheng, Yaoqi Sun et al.

Aerial-view geo-localization tends to determine an unknown position through matching the drone-view image with the geo-tagged satellite-view image. This task is mostly regarded as an image retrieval problem. The key underpinning this task is to design a series of deep neural networks to learn discriminative image descriptors. However, existing methods meet large performance drops under realistic weather, such as rain and fog, since they do not take the domain shift between the training data and multiple test environments into consideration. To minor this domain gap, we propose a Multiple-environment Self-adaptive Network (MuSe-Net) to dynamically adjust the domain shift caused by environmental changing. In particular, MuSe-Net employs a two-branch neural network containing one multiple-environment style extraction network and one self-adaptive feature extraction network. As the name implies, the multiple-environment style extraction network is to extract the environment-related style information, while the self-adaptive feature extraction network utilizes an adaptive modulation module to dynamically minimize the environment-related style gap. Extensive experiments on two widely-used benchmarks, i.e., University-1652 and CVUSA, demonstrate that the proposed MuSe-Net achieves a competitive result for geo-localization in multiple environments. Furthermore, we observe that the proposed method also shows great potential to the unseen extreme weather, such as mixing the fog, rain and snow.

Mu Chen, Zhedong Zheng, Yi Yang et al.

Unsupervised Domain Adaptation (UDA) aims to enhance the generalization of the learned model to other domains. The domain-invariant knowledge is transferred from the model trained on labeled source domain, e.g., video game, to unlabeled target domains, e.g., real-world scenarios, saving annotation expenses. Existing UDA methods for semantic segmentation usually focus on minimizing the inter-domain discrepancy of various levels, e.g., pixels, features, and predictions, for extracting domain-invariant knowledge. However, the primary intra-domain knowledge, such as context correlation inside an image, remains underexplored. In an attempt to fill this gap, we propose a unified pixel- and patch-wise self-supervised learning framework, called PiPa, for domain adaptive semantic segmentation that facilitates intra-image pixel-wise correlations and patch-wise semantic consistency against different contexts. The proposed framework exploits the inherent structures of intra-domain images, which: (1) explicitly encourages learning the discriminative pixel-wise features with intra-class compactness and inter-class separability, and (2) motivates the robust feature learning of the identical patch against different contexts or fluctuations. Extensive experiments verify the effectiveness of the proposed method, which obtains competitive accuracy on the two widely-used UDA benchmarks, i.e., 75.6 mIoU on GTA to Cityscapes and 68.2 mIoU on Synthia to Cityscapes. Moreover, our method is compatible with other UDA approaches to further improve the performance without introducing extra parameters.

Zhe Wang, Petar Veličković, Daniel Hennes et al.

Identifying key patterns of tactics implemented by rival teams, and developing effective responses, lies at the heart of modern football. However, doing so algorithmically remains an open research challenge. To address this unmet need, we propose TacticAI, an AI football tactics assistant developed and evaluated in close collaboration with domain experts from Liverpool FC. We focus on analysing corner kicks, as they offer coaches the most direct opportunities for interventions and improvements. TacticAI incorporates both a predictive and a generative component, allowing the coaches to effectively sample and explore alternative player setups for each corner kick routine and to select those with the highest predicted likelihood of success. We validate TacticAI on a number of relevant benchmark tasks: predicting receivers and shot attempts and recommending player position adjustments. The utility of TacticAI is validated by a qualitative study conducted with football domain experts at Liverpool FC. We show that TacticAI's model suggestions are not only indistinguishable from real tactics, but also favoured over existing tactics 90% of the time, and that TacticAI offers an effective corner kick retrieval system. TacticAI achieves these results despite the limited availability of gold-standard data, achieving data efficiency through geometric deep learning.

Wenhao Wang, Yifan Sun, Zongxin Yang et al.

Visual diffusion models have revolutionized the field of creative AI, producing high-quality and diverse content. However, they inevitably memorize training images or videos, subsequently replicating their concepts, content, or styles during inference. This phenomenon raises significant concerns about privacy, security, and copyright within generated outputs. In this survey, we provide the first comprehensive review of replication in visual diffusion models, marking a novel contribution to the field by systematically categorizing the existing studies into unveiling, understanding, and mitigating this phenomenon. Specifically, unveiling mainly refers to the methods used to detect replication instances. Understanding involves analyzing the underlying mechanisms and factors that contribute to this phenomenon. Mitigation focuses on developing strategies to reduce or eliminate replication. Beyond these aspects, we also review papers focusing on its real-world influence. For instance, in the context of healthcare, replication is critically worrying due to privacy concerns related to patient data. Finally, the paper concludes with a discussion of the ongoing challenges, such as the difficulty in detecting and benchmarking replication, and outlines future directions including the development of more robust mitigation techniques. By synthesizing insights from diverse studies, this paper aims to equip researchers and practitioners with a deeper understanding at the intersection between AI technology and social good. We release this project at https://github.com/WangWenhao0716/Awesome-Diffusion-Replication.

Hehe Fan, Xin Yu, Yuhang Ding et al.

Point cloud sequences are irregular and unordered in the spatial dimension while exhibiting regularities and order in the temporal dimension. Therefore, existing grid based convolutions for conventional video processing cannot be directly applied to spatio-temporal modeling of raw point cloud sequences. In this paper, we propose a point spatio-temporal (PST) convolution to achieve informative representations of point cloud sequences. The proposed PST convolution first disentangles space and time in point cloud sequences. Then, a spatial convolution is employed to capture the local structure of points in the 3D space, and a temporal convolution is used to model the dynamics of the spatial regions along the time dimension. Furthermore, we incorporate the proposed PST convolution into a deep network, namely PSTNet, to extract features of point cloud sequences in a hierarchical manner. Extensive experiments on widely-used 3D action recognition and 4D semantic segmentation datasets demonstrate the effectiveness of PSTNet to model point cloud sequences.

Dewei Zhou, Xinyu Huang, Xun Wang et al.

Generative visual models fundamentally struggle with precise spatial control. This arises from a core disconnect: models can process textual descriptions of space but cannot directly map numerical coordinates onto the 2D image canvas. We introduce MetaPoint, a method that bridges this gap by representing a continuous 2D coordinate as a single, special token. Crucially, MetaPoint requires no new architectural components; it directly leverages the model's inherent positional encoding schemes to interpret these coordinates, treating our token as a virtual point on the canvas. This lightweight approach enables pixel-level control of an object's position with one token or its bounding box with two, all without requiring architectural changes or bespoke attention masking. The MetaPoint tokens are designed to be compositional, serving as spatial primitives. This allows a planner agent to decompose a high-level user request into a structured sequence of primitives for the generator. By providing a simple, precise, and scalable building block for spatial control, MetaPoint unlocks more powerful compositional generative agents and enables intuitive, interactive editing systems.

Rui Liu, Xiaohan Wang, Wenguan Wang et al.

Vision-language navigation (VLN), which entails an agent to navigate 3D environments following human instructions, has shown great advances. However, current agents are built upon panoramic observations, which hinders their ability to perceive 3D scene geometry and easily leads to ambiguous selection of panoramic view. To address these limitations, we present a BEV Scene Graph (BSG), which leverages multi-step BEV representations to encode scene layouts and geometric cues of indoor environment under the supervision of 3D detection. During navigation, BSG builds a local BEV representation at each step and maintains a BEV-based global scene map, which stores and organizes all the online collected local BEV representations according to their topological relations. Based on BSG, the agent predicts a local BEV grid-level decision score and a global graph-level decision score, combined with a sub-view selection score on panoramic views, for more accurate action prediction. Our approach significantly outperforms state-of-the-art methods on REVERIE, R2R, and R4R, showing the potential of BEV perception in VLN.

Yuan Gan, Zongxin Yang, Xihang Yue et al.

Audio-driven talking-head synthesis is a popular research topic for virtual human-related applications. However, the inflexibility and inefficiency of existing methods, which necessitate expensive end-to-end training to transfer emotions from guidance videos to talking-head predictions, are significant limitations. In this work, we propose the Emotional Adaptation for Audio-driven Talking-head (EAT) method, which transforms emotion-agnostic talking-head models into emotion-controllable ones in a cost-effective and efficient manner through parameter-efficient adaptations. Our approach utilizes a pretrained emotion-agnostic talking-head transformer and introduces three lightweight adaptations (the Deep Emotional Prompts, Emotional Deformation Network, and Emotional Adaptation Module) from different perspectives to enable precise and realistic emotion controls. Our experiments demonstrate that our approach achieves state-of-the-art performance on widely-used benchmarks, including LRW and MEAD. Additionally, our parameter-efficient adaptations exhibit remarkable generalization ability, even in scenarios where emotional training videos are scarce or nonexistent. Project website: https://yuangan.github.io/eat/

Difei Gao, Luowei Zhou, Lei Ji et al.

To build Video Question Answering (VideoQA) systems capable of assisting humans in daily activities, seeking answers from long-form videos with diverse and complex events is a must. Existing multi-modal VQA models achieve promising performance on images or short video clips, especially with the recent success of large-scale multi-modal pre-training. However, when extending these methods to long-form videos, new challenges arise. On the one hand, using a dense video sampling strategy is computationally prohibitive. On the other hand, methods relying on sparse sampling struggle in scenarios where multi-event and multi-granularity visual reasoning are required. In this work, we introduce a new model named Multi-modal Iterative Spatial-temporal Transformer (MIST) to better adapt pre-trained models for long-form VideoQA. Specifically, MIST decomposes traditional dense spatial-temporal self-attention into cascaded segment and region selection modules that adaptively select frames and image regions that are closely relevant to the question itself. Visual concepts at different granularities are then processed efficiently through an attention module. In addition, MIST iteratively conducts selection and attention over multiple layers to support reasoning over multiple events. The experimental results on four VideoQA datasets, including AGQA, NExT-QA, STAR, and Env-QA, show that MIST achieves state-of-the-art performance and is superior at computation efficiency and interpretability.

Lin Li, Jun Xiao, Hanrong Shi et al.

Nearly all existing scene graph generation (SGG) models have overlooked the ground-truth annotation qualities of mainstream SGG datasets, i.e., they assume: 1) all the manually annotated positive samples are equally correct; 2) all the un-annotated negative samples are absolutely background. In this paper, we argue that neither of the assumptions applies to SGG: there are numerous noisy ground-truth predicate labels that break these two assumptions and harm the training of unbiased SGG models. To this end, we propose a novel NoIsy label CorrEction and Sample Training strategy for SGG: NICEST. Specifically, it consists of two parts: NICE and NIST, which rule out these noisy label issues by generating high-quality samples and the effective training strategy, respectively. NICE first detects noisy samples and then reassigns them more high-quality soft predicate labels. NIST is a multi-teacher knowledge distillation based training strategy, which enables the model to learn unbiased fusion knowledge. And a dynamic trade-off weighting strategy in NIST is designed to penalize the bias of different teachers. Due to the model-agnostic nature of both NICE and NIST, our NICEST can be seamlessly incorporated into any SGG architecture to boost its performance on different predicate categories. In addition, to better evaluate the generalization of SGG models, we further propose a new benchmark VG-OOD, by re-organizing the prevalent VG dataset and deliberately making the predicate distributions of the training and test sets as different as possible for each subject-object category pair. This new benchmark helps disentangle the influence of subject-object category based frequency biases. Extensive ablations and results on different backbones and tasks have attested to the effectiveness and generalization ability of each component of NICEST.

Dianyi Yang, Jiadong Tang, Yu Gao et al.

Fisheye cameras suffer from image distortion while having a large field of view(LFOV). And this fact leads to poor performance on some fisheye vision tasks. One of the solutions is to optimize the current vision algorithm for fisheye images. However, most of the CNN-based methods and the Transformer-based methods lack the capability of leveraging distortion information efficiently. In this work, we propose a novel patch embedding method called Sector Patch Embedding(SPE), conforming to the distortion pattern of the fisheye image. Furthermore, we put forward a synthetic fisheye dataset based on the ImageNet-1K and explore the performance of several Transformer models on the dataset. The classification top-1 accuracy of ViT and PVT is improved by 0.75% and 2.8% with SPE respectively. The experiments show that the proposed sector patch embedding method can better perceive distortion and extract features on the fisheye images. Our method can be easily adopted to other Transformer-based models. Source code is at https://github.com/IN2-ViAUn/Sector-Patch-Embedding.

Yi Yang, Victor G. Lopez, Matthias A. Müller

In this paper, we propose a moving horizon estimation (MHE)-based training method for feedforward neural networks (FNNs) with rectified linear unit (ReLU) activation functions to determine their ideal weights from a control-theoretic perspective. This allows for a rigorous theoretical analysis of the trained network. First, we reformulate the FNN as a dynamical system with the weights as states. Then, we investigate the local observability of such a system. For two-layer FNNs with fixed output weights, we derive a sufficient condition under which the observability rank condition holds, ensuring a locally observable state. We also show that multi-layer FNNs in general fail to satisfy the observability rank condition. Based on this analysis, we develop a persistently exciting (PE) input design method, which renders a state distinguishable from its neighbors. The resulting local observability provides convergence guarantees for the proposed MHE-based training, where only the projection of the state onto the observable subspace is updated using a fixed-length window of input-output data. The effectiveness of the approach is illustrated via numerical examples.

Xiaohan Wang, Wenguan Wang, Jiayi Shao et al.

Recently, visual-language navigation (VLN) -- entailing robot agents to follow navigation instructions -- has shown great advance. However, existing literature put most emphasis on interpreting instructions into actions, only delivering "dumb" wayfinding agents. In this article, we devise LANA, a language-capable navigation agent which is able to not only execute human-written navigation commands, but also provide route descriptions to humans. This is achieved by simultaneously learning instruction following and generation with only one single model. More specifically, two encoders, respectively for route and language encoding, are built and shared by two decoders, respectively, for action prediction and instruction generation, so as to exploit cross-task knowledge and capture task-specific characteristics. Throughout pretraining and fine-tuning, both instruction following and generation are set as optimization objectives. We empirically verify that, compared with recent advanced task-specific solutions, LANA attains better performances on both instruction following and route description, with nearly half complexity. In addition, endowed with language generation capability, LANA can explain to humans its behaviors and assist human's wayfinding. This work is expected to foster future efforts towards building more trustworthy and socially-intelligent navigation robots.

Liulei Li, Tianfei Zhou, Wenguan Wang et al.

Our target is to learn visual correspondence from unlabeled videos. We develop LIIR, a locality-aware inter-and intra-video reconstruction framework that fills in three missing pieces, i.e., instance discrimination, location awareness, and spatial compactness, of self-supervised correspondence learning puzzle. First, instead of most existing efforts focusing on intra-video self-supervision only, we exploit cross video affinities as extra negative samples within a unified, inter-and intra-video reconstruction scheme. This enables instance discriminative representation learning by contrasting desired intra-video pixel association against negative inter-video correspondence. Second, we merge position information into correspondence matching, and design a position shifting strategy to remove the side-effect of position encoding during inter-video affinity computation, making our LIIR location-sensitive. Third, to make full use of the spatial continuity nature of video data, we impose a compactness-based constraint on correspondence matching, yielding more sparse and reliable solutions. The learned representation surpasses self-supervised state-of-the-arts on label propagation tasks including objects, semantic parts, and keypoints.

Yaowei Li, Ruijie Quan, Linchao Zhu et al.

Large-scale pre-training has brought unimodal fields such as computer vision and natural language processing to a new era. Following this trend, the size of multi-modal learning models constantly increases, leading to an urgent need to reduce the massive computational cost of finetuning these models for downstream tasks. In this paper, we propose an efficient and flexible multimodal fusion method, namely PMF, tailored for fusing unimodally pre-trained transformers. Specifically, we first present a modular multimodal fusion framework that exhibits high flexibility and facilitates mutual interactions among different modalities. In addition, we disentangle vanilla prompts into three types in order to learn different optimizing objectives for multimodal learning. It is also worth noting that we propose to add prompt vectors only on the deep layers of the unimodal transformers, thus significantly reducing the training memory usage. Experiment results show that our proposed method achieves comparable performance to several other multimodal finetuning methods with less than 3% trainable parameters and up to 66% saving of training memory usage.

Wenguan Wang, Yi Yang, Fei Wu

Neural-symbolic computing (NeSy), which pursues the integration of the symbolic and statistical paradigms of cognition, has been an active research area of Artificial Intelligence (AI) for many years. As NeSy shows promise of reconciling the advantages of reasoning and interpretability of symbolic representation and robust learning in neural networks, it may serve as a catalyst for the next generation of AI. In the present paper, we provide a systematic overview of the recent developments and important contributions of NeSy research. Firstly, we introduce study history of this area, covering early work and foundations. We further discuss background concepts and identify key driving factors behind the development of NeSy. Afterward, we categorize recent landmark approaches along several main characteristics that underline this research paradigm, including neural-symbolic integration, knowledge representation, knowledge embedding, and functionality. Next, we briefly discuss the successful application of modern NeSy approaches in several domains. Then, we benchmark several NeSy methods on three representative application tasks. Finally, we identify the open problems together with potential future research directions. This survey is expected to help new researchers enter this rapidly evolving field and accelerate the progress towards data-and knowledge-driven AI.