Xuelong Li

Xiang Li, Yiqun Yao, Xin Jiang et al.

Large Language Models (LLMs) represent a significant stride toward Artificial General Intelligence. As scaling laws underscore the potential of increasing model sizes, the academic community has intensified its investigations into LLMs with capacities exceeding 50 billion parameters. This technical report builds on our prior work with Tele-FLM (also known as FLM-2), a publicly available 52-billion-parameter model. We delve into two primary areas: we first discuss our observation of Supervised Fine-tuning (SFT) on Tele-FLM-52B, which supports the "less is more" approach for SFT data construction; second, we demonstrate our experiments and analyses on the best practices for progressively growing a model from 52 billion to 102 billion, and subsequently to 1 trillion parameters. We will open-source a 1T model checkpoint, namely Tele-FLM-1T, to advance further training and research.

Kehui Liu, Zixin Tang, Dong Wang et al.

Leveraging the powerful reasoning capabilities of large language models (LLMs), recent LLM-based robot task planning methods yield promising results. However, they mainly focus on single or multiple homogeneous robots on simple tasks. Practically, complex long-horizon tasks always require collaboration among multiple heterogeneous robots especially with more complex action spaces, which makes these tasks more challenging. To this end, we propose COHERENT, a novel LLM-based task planning framework for collaboration of heterogeneous multi-robot systems including quadrotors, robotic dogs, and robotic arms. Specifically, a Proposal-Execution-Feedback-Adjustment (PEFA) mechanism is designed to decompose and assign actions for individual robots, where a centralized task assigner makes a task planning proposal to decompose the complex task into subtasks, and then assigns subtasks to robot executors. Each robot executor selects a feasible action to implement the assigned subtask and reports self-reflection feedback to the task assigner for plan adjustment. The PEFA loops until the task is completed. Moreover, we create a challenging heterogeneous multi-robot task planning benchmark encompassing 100 complex long-horizon tasks. The experimental results show that our work surpasses the previous methods by a large margin in terms of success rate and execution efficiency. The experimental videos, code, and benchmark are released at https://github.com/MrKeee/COHERENT.

Weichuang Li, Longhao Zhang, Dong Wang et al.

Talking head generation aims to generate faces that maintain the identity information of the source image and imitate the motion of the driving image. Most pioneering methods rely primarily on 2D representations and thus will inevitably suffer from face distortion when large head rotations are encountered. Recent works instead employ explicit 3D structural representations or implicit neural rendering to improve performance under large pose changes. Nevertheless, the fidelity of identity and expression is not so desirable, especially for novel-view synthesis. In this paper, we propose HiDe-NeRF, which achieves high-fidelity and free-view talking-head synthesis. Drawing on the recently proposed Deformable Neural Radiance Fields, HiDe-NeRF represents the 3D dynamic scene into a canonical appearance field and an implicit deformation field, where the former comprises the canonical source face and the latter models the driving pose and expression. In particular, we improve fidelity from two aspects: (i) to enhance identity expressiveness, we design a generalized appearance module that leverages multi-scale volume features to preserve face shape and details; (ii) to improve expression preciseness, we propose a lightweight deformation module that explicitly decouples the pose and expression to enable precise expression modeling. Extensive experiments demonstrate that our proposed approach can generate better results than previous works. Project page: https://www.waytron.net/hidenerf/

Chi Yan, Delin Qu, Dan Xu et al.

In this paper, we introduce \textbf{GS-SLAM} that first utilizes 3D Gaussian representation in the Simultaneous Localization and Mapping (SLAM) system. It facilitates a better balance between efficiency and accuracy. Compared to recent SLAM methods employing neural implicit representations, our method utilizes a real-time differentiable splatting rendering pipeline that offers significant speedup to map optimization and RGB-D rendering. Specifically, we propose an adaptive expansion strategy that adds new or deletes noisy 3D Gaussians in order to efficiently reconstruct new observed scene geometry and improve the mapping of previously observed areas. This strategy is essential to extend 3D Gaussian representation to reconstruct the whole scene rather than synthesize a static object in existing methods. Moreover, in the pose tracking process, an effective coarse-to-fine technique is designed to select reliable 3D Gaussian representations to optimize camera pose, resulting in runtime reduction and robust estimation. Our method achieves competitive performance compared with existing state-of-the-art real-time methods on the Replica, TUM-RGBD datasets. Project page: https://gs-slam.github.io/.

Hanqing Sun, Yanwei Pang, Jiale Cao et al.

Transformers have shown promising progress in various visual object detection tasks, including monocular 2D/3D detection and surround-view 3D detection. More importantly, the attention mechanism in the Transformer model and the 3D information extraction in binocular stereo are both similarity-based. However, directly applying existing Transformer-based detectors to binocular stereo 3D object detection leads to slow convergence and significant precision drops. We argue that a key cause of that defect is that existing Transformers ignore the binocular-stereo-specific image correspondence information. In this paper, we explore the model design of Transformers in binocular 3D object detection, focusing particularly on extracting and encoding task-specific image correspondence information. To achieve this goal, we present TS3D, a Transformer-based Stereo-aware 3D object detector. In the TS3D, a Disparity-Aware Positional Encoding (DAPE) module is proposed to embed the image correspondence information into stereo features. The correspondence is encoded as normalized sub-pixel-level disparity and is used in conjunction with sinusoidal 2D positional encoding to provide the 3D location information of the scene. To enrich multi-scale stereo features, we propose a Stereo Preserving Feature Pyramid Network (SPFPN). The SPFPN is designed to preserve the correspondence information while fusing intra-scale and aggregating cross-scale stereo features. Our proposed TS3D achieves a 41.29% Moderate Car detection average precision on the KITTI test set and takes 88 ms to detect objects from each binocular image pair. It is competitive with advanced counterparts in terms of both precision and inference speed.

Xuelong Li, Guanlin Li, Bin Zhao

Due to inadequate energy captured by the hyperspectral camera sensor in poor illumination conditions, low-light hyperspectral images (HSIs) usually suffer from low visibility, spectral distortion, and various noises. A range of HSI restoration methods have been developed, yet their effectiveness in enhancing low-light HSIs is constrained. This work focuses on the low-light HSI enhancement task, which aims to reveal the spatial-spectral information hidden in darkened areas. To facilitate the development of low-light HSI processing, we collect a low-light HSI (LHSI) dataset of both indoor and outdoor scenes. Based on Laplacian pyramid decomposition and reconstruction, we developed an end-to-end data-driven low-light HSI enhancement (HSIE) approach trained on the LHSI dataset. With the observation that illumination is related to the low-frequency component of HSI, while textural details are closely correlated to the high-frequency component, the proposed HSIE is designed to have two branches. The illumination enhancement branch is adopted to enlighten the low-frequency component with reduced resolution. The high-frequency refinement branch is utilized for refining the high-frequency component via a predicted mask. In addition, to improve information flow and boost performance, we introduce an effective channel attention block (CAB) with residual dense connection, which served as the basic block of the illumination enhancement branch. The effectiveness and efficiency of HSIE both in quantitative assessment measures and visual effects are demonstrated by experimental results on the LHSI dataset. According to the classification performance on the remote sensing Indian Pines dataset, downstream tasks benefit from the enhanced HSI. Datasets and codes are available: \href{https://github.com/guanguanboy/HSIE}{https://github.com/guanguanboy/HSIE}.

Yuheng Lei, Sitong Mao, Shunbo Zhou et al.

A generalist agent must continuously learn and adapt throughout its lifetime, achieving efficient forward transfer while minimizing catastrophic forgetting. Previous work within the dominant pretrain-then-finetune paradigm has explored parameter-efficient fine-tuning for single-task adaptation, effectively steering a frozen pretrained model with a small number of parameters. However, in the context of lifelong learning, these methods rely on the impractical assumption of a test-time task identifier and restrict knowledge sharing among isolated adapters. To address these limitations, we propose Dynamic Mixture of Progressive Parameter-Efficient Expert Library (DMPEL) for lifelong robot learning. DMPEL progressively builds a low-rank expert library and employs a lightweight router to dynamically combine experts into an end-to-end policy, enabling flexible and efficient lifelong forward transfer. Furthermore, by leveraging the modular structure of the fine-tuned parameters, we introduce expert coefficient replay, which guides the router to accurately retrieve frozen experts for previously encountered tasks. This technique mitigates forgetting while being significantly more storage- and computation-efficient than experience replay over the entire policy. Extensive experiments on the lifelong robot learning benchmark LIBERO demonstrate that our framework outperforms state-of-the-art lifelong learning methods in success rates during continual adaptation, while utilizing minimal trainable parameters and storage.

Xiaohan Liu, Yanwei Pang, Xuebin Sun et al.

Partial scan is a common approach to accelerate Magnetic Resonance Imaging (MRI) data acquisition in both 2D and 3D settings. However, accurately reconstructing images from partial scan data (i.e., incomplete k-space matrices) remains challenging due to lack of an effectively global receptive field in both spatial and k-space domains. To address this problem, we propose the following: (1) a novel convolutional operator called Faster Fourier Convolution (FasterFC) to replace the two consecutive convolution operations typically used in convolutional neural networks (e.g., U-Net, ResNet). Based on the spectral convolution theorem in Fourier theory, FasterFC employs alternating kernels of size 1 in 3D case) in different domains to extend the dual-domain receptive field to the global and achieves faster calculation speed than traditional Fast Fourier Convolution (FFC). (2) A 2D accelerated MRI method, FasterFC-End-to-End-VarNet, which uses FasterFC to improve the sensitivity maps and reconstruction quality. (3) A multi-stage 3D accelerated MRI method called FasterFC-based Single-to-group Network (FAS-Net) that utilizes a single-to-group algorithm to guide k-space domain reconstruction, followed by FasterFC-based cascaded convolutional neural networks to expand the effective receptive field in the dual-domain. Experimental results on the fastMRI and Stanford MRI Data datasets demonstrate that FasterFC improves the quality of both 2D and 3D reconstruction. Moreover, FAS-Net, as a 3D high-resolution multi-coil (eight) accelerated MRI method, achieves superior reconstruction performance in both qualitative and quantitative results compared with state-of-the-art 2D and 3D methods.

Yihao Wang, Zhigang Wang, Bin Zhao et al.

Non-line-of-sight (NLOS) tracking has drawn increasing attention in recent years, due to its ability to detect object motion out of sight. Most previous works on NLOS tracking rely on active illumination, e.g., laser, and suffer from high cost and elaborate experimental conditions. Besides, these techniques are still far from practical application due to oversimplified settings. In contrast, we propose a purely passive method to track a person walking in an invisible room by only observing a relay wall, which is more in line with real application scenarios, e.g., security. To excavate imperceptible changes in videos of the relay wall, we introduce difference frames as an essential carrier of temporal-local motion messages. In addition, we propose PAC-Net, which consists of alternating propagation and calibration, making it capable of leveraging both dynamic and static messages on a frame-level granularity. To evaluate the proposed method, we build and publish the first dynamic passive NLOS tracking dataset, NLOS-Track, which fills the vacuum of realistic NLOS datasets. NLOS-Track contains thousands of NLOS video clips and corresponding trajectories. Both real-shot and synthetic data are included. Our codes and dataset are available at https://againstentropy.github.io/NLOS-Track/.

Guanzhou Lan, Bin Zhao, Xuelong Li

Nighttime surveillance suffers from degradation due to poor illumination and arduous human annotations. It is challengable and remains a security risk at night. Existing methods rely on multi-spectral images to perceive objects in the dark, which are troubled by low resolution and color absence. We argue that the ultimate solution for nighttime surveillance is night-to-day translation, or Night2Day, which aims to translate a surveillance scene from nighttime to the daytime while maintaining semantic consistency. To achieve this, this paper presents a Disentangled Contrastive (DiCo) learning method. Specifically, to address the poor and complex illumination in the nighttime scenes, we propose a learnable physical prior, i.e., the color invariant, which provides a stable perception of a highly dynamic night environment and can be incorporated into the learning pipeline of neural networks. Targeting the surveillance scenes, we develop a disentangled representation, which is an auxiliary pretext task that separates surveillance scenes into the foreground and background with contrastive learning. Such a strategy can extract the semantics without supervision and boost our model to achieve instance-aware translation. Finally, we incorporate all the modules above into generative adversarial networks and achieve high-fidelity translation. This paper also contributes a new surveillance dataset called NightSuR. It includes six scenes to support the study on nighttime surveillance. This dataset collects nighttime images with different properties of nighttime environments, such as flare and extreme darkness. Extensive experiments demonstrate that our method outperforms existing works significantly. The dataset and source code will be released on GitHub soon.

Hongyuan Zhang, Yanchen Xu, Sida Huang et al.

Inspired by the idea of Positive-incentive Noise (Pi-Noise or $π$-Noise) that aims at learning the reliable noise beneficial to tasks, we scientifically investigate the connection between contrastive learning and $π$-noise in this paper. By converting the contrastive loss to an auxiliary Gaussian distribution to quantitatively measure the difficulty of the specific contrastive model under the information theory framework, we properly define the task entropy, the core concept of $π$-noise, of contrastive learning. It is further proved that the predefined data augmentation in the standard contrastive learning paradigm can be regarded as a kind of point estimation of $π$-noise. Inspired by the theoretical study, a framework that develops a $π$-noise generator to learn the beneficial noise (instead of estimation) as data augmentations for contrast is proposed. The designed framework can be applied to diverse types of data and is also completely compatible with the existing contrastive models. From the visualization, we surprisingly find that the proposed method successfully learns effective augmentations.

Delin Qu, Chi Yan, Dong Wang et al.

Implicit neural SLAM has achieved remarkable progress recently. Nevertheless, existing methods face significant challenges in non-ideal scenarios, such as motion blur or lighting variation, which often leads to issues like convergence failures, localization drifts, and distorted mapping. To address these challenges, we propose EN-SLAM, the first event-RGBD implicit neural SLAM framework, which effectively leverages the high rate and high dynamic range advantages of event data for tracking and mapping. Specifically, EN-SLAM proposes a differentiable CRF (Camera Response Function) rendering technique to generate distinct RGB and event camera data via a shared radiance field, which is optimized by learning a unified implicit representation with the captured event and RGBD supervision. Moreover, based on the temporal difference property of events, we propose a temporal aggregating optimization strategy for the event joint tracking and global bundle adjustment, capitalizing on the consecutive difference constraints of events, significantly enhancing tracking accuracy and robustness. Finally, we construct the simulated dataset DEV-Indoors and real captured dataset DEV-Reals containing 6 scenes, 17 sequences with practical motion blur and lighting changes for evaluations. Experimental results show that our method outperforms the SOTA methods in both tracking ATE and mapping ACC with a real-time 17 FPS in various challenging environments. Project page: https://delinqu.github.io/EN-SLAM.

Jingxian Lu, Wenke Xia, Dong Wang et al.

Online Imitation Learning struggles with the gap between extensive online exploration space and limited expert trajectories, hindering efficient exploration due to inaccurate reward estimation. Inspired by the findings from cognitive neuroscience, we hypothesize that an agent could estimate precise task-aware reward for efficient online exploration, through decomposing the target task into the objectives of "what to do" and the mechanisms of "how to do". In this work, we introduce the hybrid Key-state guided Online Imitation (KOI) learning method, which leverages the integration of semantic and motion key states as guidance for reward estimation. Initially, we utilize visual-language models to extract semantic key states from expert trajectory, indicating the objectives of "what to do". Within the intervals between semantic key states, optical flow is employed to capture motion key states to understand the mechanisms of "how to do". By integrating a thorough grasp of hybrid key states, we refine the trajectory-matching reward computation, accelerating online imitation learning with task-aware exploration. We evaluate not only the success rate of the tasks in the Meta-World and LIBERO environments, but also the trend of variance during online imitation learning, proving that our method is more sample efficient. We also conduct real-world robotic manipulation experiments to validate the efficacy of our method, demonstrating the practical applicability of our KOI method. Videos and code are available at https://gewu-lab.github.io/Keystate_Online_Imitation/.

Guanlin Li, Ke Zhang, Ting Wang et al.

Despite the impressive advancements made in recent low-light image enhancement techniques, the scarcity of paired data has emerged as a significant obstacle to further advancements. This work proposes a mean-teacher-based semi-supervised low-light enhancement (Semi-LLIE) framework that integrates the unpaired data into model training. The mean-teacher technique is a prominent semi-supervised learning method, successfully adopted for addressing high-level and low-level vision tasks. However, two primary issues hinder the naive mean-teacher method from attaining optimal performance in low-light image enhancement. Firstly, pixel-wise consistency loss is insufficient for transferring realistic illumination distribution from the teacher to the student model, which results in color cast in the enhanced images. Secondly, cutting-edge image enhancement approaches fail to effectively cooperate with the mean-teacher framework to restore detailed information in dark areas due to their tendency to overlook modeling structured information within local regions. To mitigate the above issues, we first introduce a semantic-aware contrastive loss to faithfully transfer the illumination distribution, contributing to enhancing images with natural colors. Then, we design a Mamba-based low-light image enhancement backbone to effectively enhance Mamba's local region pixel relationship representation ability with a multi-scale feature learning scheme, facilitating the generation of images with rich textural details. Further, we propose novel perceptive loss based on the large-scale vision-language Recognize Anything Model (RAM) to help generate enhanced images with richer textual details. The experimental results indicate that our Semi-LLIE surpasses existing methods in both quantitative and qualitative metrics.

Wenzhe Tian, Haijin Zeng, Yin-Ping Zhao et al.

Snapshot Compressive Imaging (SCI) enables fast spectral imaging but requires effective decoding algorithms for hyperspectral image (HSI) reconstruction from compressed measurements. Current CNN-based methods are limited in modeling long-range dependencies, while Transformer-based models face high computational complexity. Although recent Mamba models outperform CNNs and Transformers in RGB tasks concerning computational efficiency or accuracy, they are not specifically optimized to fully leverage the local spatial and spectral correlations inherent in HSIs. To address this, we propose the Cross-Scanning Mamba, named CS-Mamba, that employs a Spatial-Spectral SSM for global-local balanced context encoding and cross-channel interaction promotion. Besides, while current reconstruction algorithms perform increasingly well in simulation scenarios, they exhibit suboptimal performance on real data due to limited generalization capability. During the training process, the model may not capture the inherent features of the images but rather learn the parameters to mitigate specific noise and loss, which may lead to a decline in reconstruction quality when faced with real scenes. To overcome this challenge, we propose a masked training method to enhance the generalization ability of models. Experiment results show that our CS-Mamba achieves state-of-the-art performance and the masked training method can better reconstruct smooth features to improve the visual quality.

Yanchen Xu, Siqi Huang, Hongyuan Zhang et al.

Graph contrastive learning (GCL) has been widely used as an effective self-supervised learning method for graph representation learning. However, how to apply adequate and stable graph augmentation to generating proper views for contrastive learning remains an essential problem. Dropping edges is a primary augmentation in GCL while adding edges is not a common method due to its unstable performance. To our best knowledge, there is no theoretical analysis to study why dropping edges usually outperforms adding edges. To answer this question, we introduce a new metric, namely Error Passing Rate (EPR), to quantify how a graph fits the network. Inspired by the theoretical conclusions and the idea of positive-incentive noise, we propose a novel GCL algorithm, Error-PAssing-based Graph Contrastive Learning (EPAGCL), which uses both edge adding and edge dropping as its augmentations. To be specific, we generate views by adding and dropping edges based on the weights derived from EPR. Extensive experiments on various real-world datasets are conducted to validate the correctness of our theoretical analysis and the effectiveness of our proposed algorithm. Our code is available at: https://github.com/hyzhang98/EPAGCL.

Yiwen Tang, Ray Zhang, Jiaming Liu et al.

Large foundation models have recently emerged as a prominent focus of interest, attaining superior performance in widespread scenarios. Due to the scarcity of 3D data, many efforts have been made to adapt pre-trained transformers from vision to 3D domains. However, such 2D-to-3D approaches are still limited, due to the potential loss of spatial geometries and high computation cost. More importantly, their frameworks are mainly designed for 2D models, lacking a general any-to-3D paradigm. In this paper, we introduce Any2Point, a parameter-efficient method to empower any-modality large models (vision, language, audio) for 3D understanding. Given a frozen transformer from any source modality, we propose a 3D-to-any (1D or 2D) virtual projection strategy that correlates the input 3D points to the original 1D or 2D positions within the source modality. This mechanism enables us to assign each 3D token with a positional encoding paired with the pre-trained model, which avoids 3D geometry loss caused by the true projection and better motivates the transformer for 3D learning with 1D/2D positional priors. Then, within each transformer block, we insert an any-to-3D guided adapter module for parameter-efficient fine-tuning. The adapter incorporates prior spatial knowledge from the source modality to guide the local feature aggregation of 3D tokens, compelling the semantic adaption of any-modality transformers. We conduct extensive experiments to showcase the effectiveness and efficiency of our method. Code and models are released at https://github.com/Ivan-Tang-3D/Any2Point.

Ruishu Zhu, Zhihao Huang, Jiacheng Sun et al.

Multi-view image generation from a single image and text description remains challenging due to the difficulty of maintaining geometric consistency across different viewpoints. Existing approaches typically rely on 3D-aware architectures or specialized diffusion models that require extensive multi-view training data and complex geometric priors. In this work, we introduce ViewMask-1-to-3, a pioneering approach to apply discrete diffusion models to multi-view image generation. Unlike continuous diffusion methods that operate in latent spaces, ViewMask-1-to-3 formulates multi-view synthesis as a discrete sequence modeling problem, where each viewpoint is represented as visual tokens obtained through MAGVIT-v2 tokenization. By unifying language and vision through masked token prediction, our approach enables progressive generation of multiple viewpoints through iterative token unmasking with text input. ViewMask-1-to-3 achieves cross-view consistency through simple random masking combined with self-attention, eliminating the requirement for complex 3D geometric constraints or specialized attention architectures. Our approach demonstrates that discrete diffusion provides a viable and simple alternative to existing multi-view generation methods, ranking first on average across GSO and 3D-FUTURE datasets in terms of PSNR, SSIM, and LPIPS, while maintaining architectural simplicity.

Bingyu Li, Da Zhang, Zhiyuan Zhao et al.

Multimodal semantic segmentation is a pivotal component of computer vision and typically surpasses unimodal methods by utilizing rich information set from various sources.Current models frequently adopt modality-specific frameworks that inherently biases toward certain modalities. Although these biases might be advantageous in specific situations, they generally limit the adaptability of the models across different multimodal contexts, thereby potentially impairing performance. To address this issue, we leverage the inherent capabilities of the model itself to discover the optimal equilibrium in multimodal fusion and introduce U3M: An Unbiased Multiscale Modal Fusion Model for Multimodal Semantic Segmentation. Specifically, this method involves an unbiased integration of multimodal visual data. Additionally, we employ feature fusion at multiple scales to ensure the effective extraction and integration of both global and local features. Experimental results demonstrate that our approach achieves superior performance across multiple datasets, verifing its efficacy in enhancing the robustness and versatility of semantic segmentation in diverse settings. Our code is available at U3M-multimodal-semantic-segmentation.

Hongjun An, Yifan Chen, Zhe Sun et al.

Current large language models (LLMs) primarily utilize next-token prediction method for inference, which significantly impedes their processing speed. In this paper, we introduce a novel inference methodology termed next-sentence prediction, aiming at enhancing the inference efficiency of LLMs. We present Sentence Variational Autoencoder (SentenceVAE), which includes a Sentence Encoder to compress multiple tokens in a sentence into a single token, and a Sentence Decoder to reconstruct it. By integrating SentenceVAE into the input and output layers of LLMs, we develop Sentence-level LLMs (SLLMs) that employ a sentence-by-sentence inference method. In addition, the SentenceVAE module of SLLMs can maintain the integrity of the original semantic content by segmenting the context into sentences, thereby improving accuracy while boosting inference speed. Moreover, compared to previous LLMs, SLLMs process fewer tokens over equivalent context length, significantly reducing memory demands for self-attention computation and facilitating the handling of longer context. Extensive experiments on Wanjuan dataset have revealed that the proposed method can accelerate inference speed by 204~365%, reduce perplexity (PPL) to 46~75% of its original metric, and decrease memory overhead by 86~91% for the equivalent context length, compared to previous token-by-token methods.

Zhihao Huang, Xi Qiu, Yukuo Ma et al.

Autoregressive models have achieved significant success in image generation. However, unlike the inherent hierarchical structure of image information in the spectral domain, standard autoregressive methods typically generate pixels sequentially in a fixed spatial order. To better leverage this spectral hierarchy, we introduce NextFrequency Image Generation (NFIG). NFIG is a novel framework that decomposes the image generation process into multiple frequency-guided stages. NFIG aligns the generation process with the natural image structure. It does this by first generating low-frequency components, which efficiently capture global structure with significantly fewer tokens, and then progressively adding higher-frequency details. This frequency-aware paradigm offers substantial advantages: it not only improves the quality of generated images but crucially reduces inference cost by efficiently establishing global structure early on. Extensive experiments on the ImageNet-256 benchmark validate NFIG's effectiveness, demonstrating superior performance (FID: 2.81) and a notable 1.25x speedup compared to the strong baseline VAR-d20. The source code is available at https://github.com/Pride-Huang/NFIG.

Xiang Li, Yiqun Yao, Xin Jiang et al.

Large language models (LLMs) have showcased profound capabilities in language understanding and generation, facilitating a wide array of applications. However, there is a notable paucity of detailed, open-sourced methodologies on efficiently scaling LLMs beyond 50 billion parameters with minimum trial-and-error cost and computational resources. In this report, we introduce Tele-FLM (aka FLM-2), a 52B open-sourced multilingual large language model that features a stable, efficient pre-training paradigm and enhanced factual judgment capabilities. Tele-FLM demonstrates superior multilingual language modeling abilities, measured by BPB on textual corpus. Besides, in both English and Chinese foundation model evaluation, it is comparable to strong open-sourced models that involve larger pre-training FLOPs, such as Llama2-70B and DeepSeek-67B. In addition to the model weights, we share the core designs, engineering practices, and training details, which we expect to benefit both the academic and industrial communities.

Jinda Lu, Shuo Wang, Yanbin Hao et al.

Recent adaptations can boost the low-shot capability of Contrastive Vision-Language Pre-training (CLIP) by effectively facilitating knowledge transfer. However, these adaptation methods are usually operated on the global view of an input image, and thus biased perception of partial local details of the image. To solve this problem, we propose a Visual Content Refinement (VCR) before the adaptation calculation during the test stage. Specifically, we first decompose the test image into different scales to shift the feature extractor's attention to the details of the image. Then, we select the image view with the max prediction margin in each scale to filter out the noisy image views, where the prediction margins are calculated from the pre-trained CLIP model. Finally, we merge the content of the aforementioned selected image views based on their scales to construct a new robust representation. Thus, the merged content can be directly used to help the adapter focus on both global and local parts without any extra training parameters. We apply our method to 3 popular low-shot benchmark tasks with 13 datasets and achieve a significant improvement over state-of-the-art methods. For example, compared to the baseline (Tip-Adapter) on the few-shot classification task, our method achieves about 2\% average improvement for both training-free and training-need settings.

Bingyu Li, Tao Huo, Da Zhang et al.

Accurate segmentation of marine organisms is vital for biodiversity monitoring and ecological assessment, yet existing datasets and models remain largely limited to terrestrial scenes. To bridge this gap, we introduce \textbf{AquaOV255}, the first large-scale and fine-grained underwater segmentation dataset containing 255 categories and over 20K images, covering diverse categories for open-vocabulary (OV) evaluation. Furthermore, we establish the first underwater OV segmentation benchmark, \textbf{UOVSBench}, by integrating AquaOV255 with five additional underwater datasets to enable comprehensive evaluation. Alongside, we present \textbf{Earth2Ocean}, a training-free OV segmentation framework that transfers terrestrial vision--language models (VLMs) to underwater domains without any additional underwater training. Earth2Ocean consists of two core components: a Geometric-guided Visual Mask Generator (\textbf{GMG}) that refines visual features via self-similarity geometric priors for local structure perception, and a Category-visual Semantic Alignment (\textbf{CSA}) module that enhances text embeddings through multimodal large language model reasoning and scene-aware template construction. Extensive experiments on the UOVSBench benchmark demonstrate that Earth2Ocean achieves significant performance improvement on average while maintaining efficient inference.

Yifan Chen, Yang Zhao, Xuelong Li

Gait recognition is an important recognition technology, because gait is not easy to camouflage and does not need cooperation to recognize subjects. However, many existing methods are inadequate in preserving both temporal information and fine-grained information, thus reducing its discrimination. This problem is more serious when the subjects with similar walking postures are identified. In this paper, we try to enhance the discrimination of spatio-temporal gait features from two aspects: effective extraction of spatio-temporal gait features and reasonable refinement of extracted features. Thus our method is proposed, it consists of Spatio-temporal Feature Extraction (SFE) and Global Distance Alignment (GDA). SFE uses Temporal Feature Fusion (TFF) and Fine-grained Feature Extraction (FFE) to effectively extract the spatio-temporal features from raw silhouettes. GDA uses a large number of unlabeled gait data in real life as a benchmark to refine the extracted spatio-temporal features. GDA can make the extracted features have low inter-class similarity and high intra-class similarity, thus enhancing their discrimination. Extensive experiments on mini-OUMVLP and CASIA-B have proved that we have a better result than some state-of-the-art methods.

Pingrui Zhang, Yifei Su, Pengyuan Wu et al.

Vision-and-Language Navigation (VLN) requires the agent to navigate by following natural instructions under partial observability, making it difficult to align perception with language. Recent methods mitigate this by imagining future scenes, yet they rely on vision-based synthesis, leading to high computational cost and redundant details. To this end, we propose to adaptively imagine key environmental semantics via \textit{language} form, enabling a more reliable and efficient strategy. Specifically, we introduce a novel Adaptive Text Dreamer (ATD), a dual-branch self-guided imagination policy built upon a large language model (LLM). ATD is designed with a human-like left-right brain architecture, where the left brain focuses on logical integration, and the right brain is responsible for imaginative prediction of future scenes. To achieve this, we fine-tune only the Q-former within both brains to efficiently activate domain-specific knowledge in the LLM, enabling dynamic updates of logical reasoning and imagination during navigation. Furthermore, we introduce a cross-interaction mechanism to regularize the imagined outputs and inject them into a navigation expert module, allowing ATD to jointly exploit both the reasoning capacity of the LLM and the expertise of the navigation model. We conduct extensive experiments on the R2R benchmark, where ATD achieves state-of-the-art performance with fewer parameters. The code is \href{https://github.com/zhangpingrui/Adaptive-Text-Dreamer}{here}.

Jinghui Yuan, Fangyuan Xie, Feiping Nie et al.

The indicator matrix plays an important role in machine learning, but optimizing it is an NP-hard problem. We propose a new relaxation of the indicator matrix and prove that this relaxation forms a manifold, which we call the Relaxed Indicator Matrix Manifold (RIM manifold). Based on Riemannian geometry, we develop a Riemannian toolbox for optimization on the RIM manifold. Specifically, we provide several methods of Retraction, including a fast Retraction method to obtain geodesics. We point out that the RIM manifold is a generalization of the double stochastic manifold, and it is much faster than existing methods on the double stochastic manifold, which has a complexity of \( \mathcal{O}(n^3) \), while RIM manifold optimization is \( \mathcal{O}(n) \) and often yields better results. We conducted extensive experiments, including image denoising, with millions of variables to support our conclusion, and applied the RIM manifold to Ratio Cut, we provide a rigorous convergence proof and achieve clustering results that outperform the state-of-the-art methods. Our Code in \href{https://github.com/Yuan-Jinghui/Riemannian-Optimization-on-Relaxed-Indicator-Matrix-Manifold}{here}.

Fan Zhang, Zhi-Qi Cheng, Jian Zhao et al. · cmu, uw

Semi-supervised learning has emerged as a promising approach to tackle the challenge of label scarcity in facial expression recognition (FER) task. However, current state-of-the-art methods primarily focus on one side of the coin, i.e., generating high-quality pseudo-labels, while overlooking the other side: enhancing expression-relevant representations. In this paper, we unveil both sides of the coin by proposing a unified framework termed hierarchicaL dEcoupling And Fusing (LEAF) to coordinate expression-relevant representations and pseudo-labels for semi-supervised FER. LEAF introduces a hierarchical expression-aware aggregation strategy that operates at three levels: semantic, instance, and category. (1) At the semantic and instance levels, LEAF decouples representations into expression-agnostic and expression-relevant components, and adaptively fuses them using learnable gating weights. (2) At the category level, LEAF assigns ambiguous pseudo-labels by decoupling predictions into positive and negative parts, and employs a consistency loss to ensure agreement between two augmented views of the same image. Extensive experiments on benchmark datasets demonstrate that by unveiling and harmonizing both sides of the coin, LEAF outperforms state-of-the-art semi-supervised FER methods, effectively leveraging both labeled and unlabeled data. Moreover, the proposed expression-aware aggregation strategy can be seamlessly integrated into existing semi-supervised frameworks, leading to significant performance gains. Our code is available at \url{https://github.com/zfkarl/LEAF}.

Bingyu Li, Haocheng Dong, Da Zhang et al.

Open-Vocabulary Remote Sensing Image Segmentation (OVRSIS), an emerging task that adapts Open-Vocabulary Segmentation (OVS) to the remote sensing (RS) domain, remains underexplored due to the absence of a unified evaluation benchmark and the domain gap between natural and RS images. To bridge these gaps, we first establish a standardized OVRSIS benchmark (\textbf{OVRSISBench}) based on widely-used RS segmentation datasets, enabling consistent evaluation across methods. Using this benchmark, we comprehensively evaluate several representative OVS/OVRSIS models and reveal their limitations when directly applied to remote sensing scenarios. Building on these insights, we propose \textbf{RSKT-Seg}, a novel open-vocabulary segmentation framework tailored for remote sensing. RSKT-Seg integrates three key components: (1) a Multi-Directional Cost Map Aggregation (RS-CMA) module that captures rotation-invariant visual cues by computing vision-language cosine similarities across multiple directions; (2) an Efficient Cost Map Fusion (RS-Fusion) transformer, which jointly models spatial and semantic dependencies with a lightweight dimensionality reduction strategy; and (3) a Remote Sensing Knowledge Transfer (RS-Transfer) module that injects pre-trained knowledge and facilitates domain adaptation via enhanced upsampling. Extensive experiments on the benchmark show that RSKT-Seg consistently outperforms strong OVS baselines by +3.8 mIoU and +5.9 mACC, while achieving 2x faster inference through efficient aggregation. Our code is \href{https://github.com/LiBingyu01/RSKT-Seg}{\textcolor{blue}{here}}.

X. Feng, S. Hu, X. Li et al.

Vision-language tracking aims to locate the target object in the video sequence using a template patch and a language description provided in the initial frame. To achieve robust tracking, especially in complex long-term scenarios that reflect real-world conditions as recently highlighted by MGIT, it is essential not only to characterize the target features but also to utilize the context features related to the target. However, the visual and textual target-context cues derived from the initial prompts generally align only with the initial target state. Due to their dynamic nature, target states are constantly changing, particularly in complex long-term sequences. It is intractable for these cues to continuously guide Vision-Language Trackers (VLTs). Furthermore, for the text prompts with diverse expressions, our experiments reveal that existing VLTs struggle to discern which words pertain to the target or the context, complicating the utilization of textual cues. In this work, we present a novel tracker named ATCTrack, which can obtain multimodal cues Aligned with the dynamic target states through comprehensive Target-Context feature modeling, thereby achieving robust tracking. Specifically, (1) for the visual modality, we propose an effective temporal visual target-context modeling approach that provides the tracker with timely visual cues. (2) For the textual modality, we achieve precise target words identification solely based on textual content, and design an innovative context words calibration method to adaptively utilize auxiliary context words. (3) We conduct extensive experiments on mainstream benchmarks and ATCTrack achieves a new SOTA performance. The code and models will be released at: https://github.com/XiaokunFeng/ATCTrack.

Qianqian Sun, Jixiang Luo, Dell Zhang et al.

Recent advancements in image editing have utilized large-scale multimodal models to enable intuitive, natural instruction-driven interactions. However, conventional methods still face significant challenges, particularly in spatial reasoning, precise region segmentation, and maintaining semantic consistency, especially in complex scenes. To overcome these challenges, we introduce SmartFreeEdit, a novel end-to-end framework that integrates a multimodal large language model (MLLM) with a hypergraph-enhanced inpainting architecture, enabling precise, mask-free image editing guided exclusively by natural language instructions. The key innovations of SmartFreeEdit include:(1)the introduction of region aware tokens and a mask embedding paradigm that enhance the spatial understanding of complex scenes;(2) a reasoning segmentation pipeline designed to optimize the generation of editing masks based on natural language instructions;and (3) a hypergraph-augmented inpainting module that ensures the preservation of both structural integrity and semantic coherence during complex edits, overcoming the limitations of local-based image generation. Extensive experiments on the Reason-Edit benchmark demonstrate that SmartFreeEdit surpasses current state-of-the-art methods across multiple evaluation metrics, including segmentation accuracy, instruction adherence, and visual quality preservation, while addressing the issue of local information focus and improving global consistency in the edited image. Our project will be available at https://github.com/smileformylove/SmartFreeEdit.

Zhuoming Li, Aitong Liu, Mengxi Jia et al.

Free-form gesture understanding is highly appealing for human-computer interaction, as it liberates users from the constraints of predefined gesture categories. However, the sole existing solution GestureGPT suffers from limited recognition accuracy and slow response times. In this paper, we propose Gestura, an end-to-end system for free-form gesture understanding. Gestura harnesses a pre-trained Large Vision-Language Model (LVLM) to align the highly dynamic and diverse patterns of free-form gestures with high-level semantic concepts. To better capture subtle hand movements across different styles, we introduce a Landmark Processing Module that compensate for LVLMs' lack of fine-grained domain knowledge by embedding anatomical hand priors. Further, a Chain-of-Thought (CoT) reasoning strategy enables step-by-step semantic inference, transforming shallow knowledge into deep semantic understanding and significantly enhancing the model's ability to interpret ambiguous or unconventional gestures. Together, these components allow Gestura to achieve robust and adaptable free-form gesture comprehension. Additionally, we have developed the first open-source dataset for free-form gesture intention reasoning and understanding with over 300,000 annotated QA pairs.

Da Zhang, Chenggang Rong, Bingyu Li et al.

Large vision-language models (VLMs) have achieved remarkable success in natural scene understanding, yet their application to underwater environments remains largely unexplored. Underwater imagery presents unique challenges including severe light attenuation, color distortion, and suspended particle scattering, while requiring specialized knowledge of marine ecosystems and organism taxonomy. To bridge this gap, we introduce UWBench, a comprehensive benchmark specifically designed for underwater vision-language understanding. UWBench comprises 15,003 high-resolution underwater images captured across diverse aquatic environments, encompassing oceans, coral reefs, and deep-sea habitats. Each image is enriched with human-verified annotations including 15,281 object referring expressions that precisely describe marine organisms and underwater structures, and 124,983 question-answer pairs covering diverse reasoning capabilities from object recognition to ecological relationship understanding. The dataset captures rich variations in visibility, lighting conditions, and water turbidity, providing a realistic testbed for model evaluation. Based on UWBench, we establish three comprehensive benchmarks: detailed image captioning for generating ecologically informed scene descriptions, visual grounding for precise localization of marine organisms, and visual question answering for multimodal reasoning about underwater environments. Extensive experiments on state-of-the-art VLMs demonstrate that underwater understanding remains challenging, with substantial room for improvement. Our benchmark provides essential resources for advancing vision-language research in underwater contexts and supporting applications in marine science, ecological monitoring, and autonomous underwater exploration. Our code and benchmark will be available.

X. Feng, D. Zhang, S. Hu et al.

Effectively modeling and utilizing spatiotemporal features from RGB and other modalities (\eg, depth, thermal, and event data, denoted as X) is the core of RGB-X tracker design. Existing methods often employ two parallel branches to separately process the RGB and X input streams, requiring the model to simultaneously handle two dispersed feature spaces, which complicates both the model structure and computation process. More critically, intra-modality spatial modeling within each dispersed space incurs substantial computational overhead, limiting resources for inter-modality spatial modeling and temporal modeling. To address this, we propose a novel tracker, CSTrack, which focuses on modeling Compact Spatiotemporal features to achieve simple yet effective tracking. Specifically, we first introduce an innovative Spatial Compact Module that integrates the RGB-X dual input streams into a compact spatial feature, enabling thorough intra- and inter-modality spatial modeling. Additionally, we design an efficient Temporal Compact Module that compactly represents temporal features by constructing the refined target distribution heatmap. Extensive experiments validate the effectiveness of our compact spatiotemporal modeling method, with CSTrack achieving new SOTA results on mainstream RGB-X benchmarks. The code and models will be released at: https://github.com/XiaokunFeng/CSTrack.

Jia Zeng, Qingwen Bu, Bangjun Wang et al.

Representation learning approaches for robotic manipulation have boomed in recent years. Due to the scarcity of in-domain robot data, prevailing methodologies tend to leverage large-scale human video datasets to extract generalizable features for visuomotor policy learning. Despite the progress achieved, prior endeavors disregard the interactive dynamics that capture behavior patterns and physical interaction during the manipulation process, resulting in an inadequate understanding of the relationship between objects and the environment. To this end, we propose a general pre-training pipeline that learns Manipulation by Predicting the Interaction (MPI) and enhances the visual representation.Given a pair of keyframes representing the initial and final states, along with language instructions, our algorithm predicts the transition frame and detects the interaction object, respectively. These two learning objectives achieve superior comprehension towards "how-to-interact" and "where-to-interact". We conduct a comprehensive evaluation of several challenging robotic tasks.The experimental results demonstrate that MPI exhibits remarkable improvement by 10% to 64% compared with previous state-of-the-art in real-world robot platforms as well as simulation environments. Code and checkpoints are publicly shared at https://github.com/OpenDriveLab/MPI.

Tu Zheng, Hao Fang, Yi Zhang et al.

Lane detection is one of the most important tasks in self-driving. Due to various complex scenarios (e.g., severe occlusion, ambiguous lanes, etc.) and the sparse supervisory signals inherent in lane annotations, lane detection task is still challenging. Thus, it is difficult for the ordinary convolutional neural network (CNN) to train in general scenes to catch subtle lane feature from the raw image. In this paper, we present a novel module named REcurrent Feature-Shift Aggregator (RESA) to enrich lane feature after preliminary feature extraction with an ordinary CNN. RESA takes advantage of strong shape priors of lanes and captures spatial relationships of pixels across rows and columns. It shifts sliced feature map recurrently in vertical and horizontal directions and enables each pixel to gather global information. RESA can conjecture lanes accurately in challenging scenarios with weak appearance clues by aggregating sliced feature map. Moreover, we propose a Bilateral Up-Sampling Decoder that combines coarse-grained and fine-detailed features in the up-sampling stage. It can recover the low-resolution feature map into pixel-wise prediction meticulously. Our method achieves state-of-the-art results on two popular lane detection benchmarks (CULane and Tusimple). Code has been made available at: https://github.com/ZJULearning/resa.

Minghao Chen, Shuai Zhao, Haifeng Liu et al.

Recently, remarkable progress has been made in learning transferable representation across domains. Previous works in domain adaptation are majorly based on two techniques: domain-adversarial learning and self-training. However, domain-adversarial learning only aligns feature distributions between domains but does not consider whether the target features are discriminative. On the other hand, self-training utilizes the model predictions to enhance the discrimination of target features, but it is unable to explicitly align domain distributions. In order to combine the strengths of these two methods, we propose a novel method called Adversarial-Learned Loss for Domain Adaptation (ALDA). We first analyze the pseudo-label method, a typical self-training method. Nevertheless, there is a gap between pseudo-labels and the ground truth, which can cause incorrect training. Thus we introduce the confusion matrix, which is learned through an adversarial manner in ALDA, to reduce the gap and align the feature distributions. Finally, a new loss function is auto-constructed from the learned confusion matrix, which serves as the loss for unlabeled target samples. Our ALDA outperforms state-of-the-art approaches in four standard domain adaptation datasets. Our code is available at https://github.com/ZJULearning/ALDA.

Zili Liu, Tu Zheng, Guodong Xu et al.

Modern object detectors can rarely achieve short training time, fast inference speed, and high accuracy at the same time. To strike a balance among them, we propose the Training-Time-Friendly Network (TTFNet). In this work, we start with light-head, single-stage, and anchor-free designs, which enable fast inference speed. Then, we focus on shortening training time. We notice that encoding more training samples from annotated boxes plays a similar role as increasing batch size, which helps enlarge the learning rate and accelerate the training process. To this end, we introduce a novel approach using Gaussian kernels to encode training samples. Besides, we design the initiative sample weights for better information utilization. Experiments on MS COCO show that our TTFNet has great advantages in balancing training time, inference speed, and accuracy. It has reduced training time by more than seven times compared to previous real-time detectors while maintaining state-of-the-art performances. In addition, our super-fast version of TTFNet-18 and TTFNet-53 can outperform SSD300 and YOLOv3 by less than one-tenth of their training time, respectively. The code has been made available at \url{https://github.com/ZJULearning/ttfnet}.

Bin Zhao, Xuelong Li, Xiaoqiang Lu et al.

Weather Recognition plays an important role in our daily lives and many computer vision applications. However, recognizing the weather conditions from a single image remains challenging and has not been studied thoroughly. Generally, most previous works treat weather recognition as a single-label classification task, namely, determining whether an image belongs to a specific weather class or not. This treatment is not always appropriate, since more than one weather conditions may appear simultaneously in a single image. To address this problem, we make the first attempt to view weather recognition as a multi-label classification task, i.e., assigning an image more than one labels according to the displayed weather conditions. Specifically, a CNN-RNN based multi-label classification approach is proposed in this paper. The convolutional neural network (CNN) is extended with a channel-wise attention model to extract the most correlated visual features. The Recurrent Neural Network (RNN) further processes the features and excavates the dependencies among weather classes. Finally, the weather labels are predicted step by step. Besides, we construct two datasets for the weather recognition task and explore the relationships among different weather conditions. Experimental results demonstrate the superiority and effectiveness of the proposed approach. The new constructed datasets will be available at https://github.com/wzgwzg/Multi-Label-Weather-Recognition.

Sida Huang, Hongyuan Zhang, Xuelong Li

With the advancement of pre-trained vision-language (VL) models, enhancing the alignment between visual and linguistic modalities in downstream tasks has emerged as a critical challenge. Different from existing fine-tuning methods that add extra modules to these two modalities, we investigate whether the frozen model can be fine-tuned by customized noise. Our approach is motivated by the scientific study of beneficial noise, namely Positive-incentive Noise (Pi-noise or $π$-noise) , which quantitatively analyzes the impact of noise. It therefore implies a new scheme to learn beneficial noise distribution that can be employed to fine-tune VL models. Focusing on few-shot classification tasks based on CLIP, we reformulate the inference process of CLIP and apply variational inference, demonstrating how to generate $π$-noise towards visual and linguistic modalities. Then, we propose Positive-incentive Noise Injector (PiNI), which can fine-tune CLIP via injecting noise into both visual and text encoders. Since the proposed method can learn the distribution of beneficial noise, we can obtain more diverse embeddings of vision and language to better align these two modalities for specific downstream tasks within limited computational resources. We evaluate different noise incorporation approaches and network architectures of PiNI. The evaluation across 11 datasets demonstrates its effectiveness.

Lianrui Li, Dakuan Lu, Jiawei Shao et al.

We propose Self-correction Relative Policy Optimization (ScRPO), a novel reinforcement learning framework designed to enhance large language models on challenging mathematical problems by leveraging self-reflection and error correction. Our approach consists of two stages: (1) Trial-and-error learning stage: training the model with GRPO and collecting incorrect answers along with their corresponding questions in an error pool; (2) Self-correction learning stage: guiding the model to reflect on why its previous answers were wrong. Extensive experiments across multiple math reasoning benchmarks, including AIME, AMC, Olympiad, MATH-500, GSM8k, using Deepseek-Distill-Qwen-1.5B and Deepseek-Distill-Qwen-7B. The experimental results demonstrate that ScRPO consistently outperforms several post-training methods. These findings highlight ScRPO as a promising paradigm for enabling language models to self-improve on difficult tasks with limited external feedback, paving the way toward more reliable and capable AI systems.

Pingrui Zhang, Xianqiang Gao, Yuhan Wu et al.

In mobile manipulation, navigation and manipulation are often treated as separate problems, resulting in a significant gap between merely approaching an object and engaging with it effectively. Many navigation approaches primarily define success by proximity to the target, often overlooking the necessity for optimal positioning that facilitates subsequent manipulation. To address this, we introduce MoMa-Kitchen, a benchmark dataset comprising over 100k samples that provide training data for models to learn optimal final navigation positions for seamless transition to manipulation. Our dataset includes affordance-grounded floor labels collected from diverse kitchen environments, in which robotic mobile manipulators of different models attempt to grasp target objects amidst clutter. Using a fully automated pipeline, we simulate diverse real-world scenarios and generate affordance labels for optimal manipulation positions. Visual data are collected from RGB-D inputs captured by a first-person view camera mounted on the robotic arm, ensuring consistency in viewpoint during data collection. We also develop a lightweight baseline model, NavAff, for navigation affordance grounding that demonstrates promising performance on the MoMa-Kitchen benchmark. Our approach enables models to learn affordance-based final positioning that accommodates different arm types and platform heights, thereby paving the way for more robust and generalizable integration of navigation and manipulation in embodied AI. Project page: \href{https://momakitchen.github.io/}{https://momakitchen.github.io/}.

Xudong Yu, Chenjia Bai, Haoran He et al.

Sequential decision-making is desired to align with human intents and exhibit versatility across various tasks. Previous methods formulate it as a conditional generation process, utilizing return-conditioned diffusion models to directly model trajectory distributions. Nevertheless, the return-conditioned paradigm relies on pre-defined reward functions, facing challenges when applied in multi-task settings characterized by varying reward functions (versatility) and showing limited controllability concerning human preferences (alignment). In this work, we adopt multi-task preferences as a unified condition for both single- and multi-task decision-making, and propose preference representations aligned with preference labels. The learned representations are used to guide the conditional generation process of diffusion models, and we introduce an auxiliary objective to maximize the mutual information between representations and corresponding generated trajectories, improving alignment between trajectories and preferences. Extensive experiments in D4RL and Meta-World demonstrate that our method presents favorable performance in single- and multi-task scenarios, and exhibits superior alignment with preferences.

Yuanqi Yao, Siao Liu, Haoming Song et al.

Building a lifelong robot that can effectively leverage prior knowledge for continuous skill acquisition remains significantly challenging. Despite the success of experience replay and parameter-efficient methods in alleviating catastrophic forgetting problem, naively applying these methods causes a failure to leverage the shared primitives between skills. To tackle these issues, we propose Primitive Prompt Learning (PPL), to achieve lifelong robot manipulation via reusable and extensible primitives. Within our two stage learning scheme, we first learn a set of primitive prompts to represent shared primitives through multi-skills pre-training stage, where motion-aware prompts are learned to capture semantic and motion shared primitives across different skills. Secondly, when acquiring new skills in lifelong span, new prompts are appended and optimized with frozen pretrained prompts, boosting the learning via knowledge transfer from old skills to new ones. For evaluation, we construct a large-scale skill dataset and conduct extensive experiments in both simulation and real-world tasks, demonstrating PPL's superior performance over state-of-the-art methods.

Youquan Fu, Ruiyang Si, Hongfa Wang et al.

There is a high demand for audio-visual editing in video post-production and the film making field. While numerous models have explored audio and video editing, they struggle with object-level audio-visual operations. Specifically, object-level audio-visual editing requires the ability to perform object addition, replacement, and removal across both audio and visual modalities, while preserving the structural information of the source instances during the editing process. In this paper, we present \textbf{Object-AVEdit}, achieving the object-level audio-visual editing based on the inversion-regeneration paradigm. To achieve the object-level controllability during editing, we develop a word-to-sounding-object well-aligned audio generation model, bridging the gap in object-controllability between audio and current video generation models. Meanwhile, to achieve the better structural information preservation and object-level editing effect, we propose an inversion-regeneration holistically-optimized editing algorithm, ensuring both information retention during the inversion and better regeneration effect. Extensive experiments demonstrate that our editing model achieved advanced results in both audio-video object-level editing tasks with fine audio-visual semantic alignment. In addition, our developed audio generation model also achieved advanced performance. More results on our project page: https://gewu-lab.github.io/Object_AVEdit-website/.

Zhigang Wang, Yifei Su, Chenhui Li et al.

Open-vocabulary 3D scene understanding is indispensable for embodied agents. Recent works leverage pretrained vision-language models (VLMs) for object segmentation and project them to point clouds to build 3D maps. Despite progress, a point cloud is a set of unordered coordinates that requires substantial storage space and does not directly convey occupancy information or spatial relation, making existing methods inefficient for downstream tasks, e.g., path planning and complex text-based object retrieval. To address these issues, we propose Octree-Graph, a novel scene representation for open-vocabulary 3D scene understanding. Specifically, a Chronological Group-wise Segment Merging (CGSM) strategy and an Instance Feature Aggregation (IFA) algorithm are first designed to get 3D instances and corresponding semantic features. Subsequently, an adaptive-octree structure is developed that stores semantics and depicts the occupancy of an object adjustably according to its shape. Finally, the Octree-Graph is constructed where each adaptive-octree acts as a graph node, and edges describe the spatial relations among nodes. Extensive experiments on various tasks are conducted on several widely-used datasets, demonstrating the versatility and effectiveness of our method.

Xiaozhen Qiao, Peng Huang, Jiakang Yuan et al.

Test-time adaptation (TTA) is crucial in maintaining performance of Vision Language Models (VLMs) when facing distribution shifts, particularly when the source data or target labels are inaccessible. Existing TTA methods predominantly leverage the output probability distribution of CLIP for feature evaluation, resulting in biases under domain shifts, which cause misclassified features due to text priors or incorrect textual associations. To address these issues, we propose \underline{B}idirectional Prototype-Reward co-Evolution (BPRE), a novel VLMs framework with TTA that integrates feature quality assessment with prototype evolution via a synergistic feedback loop. First, the Multi-dimensional Quality-aware Reward Module (MQRM) is designed to evaluate feature quality and guide prototype refinement precisely. The continuous refinement of prototype quality via Prototype-Reward Interactive Evolution (PRIE) enhances the computation more robust. Through this bidirectional interaction, the precision of rewards and prototype evolution mutually reinforce each other, forming a self-evolving feedback cycle. Extensive experiments conducted on 15 diverse recognition datasets demonstrate that our model consistently achieves superior performance compared to other SOTA methods, and advances VLM generalization capabilities through emphasizing comprehensive feature evaluation.

Xunzhi Xiang, Yabo Chen, Guiyu Zhang et al.

Current autoregressive diffusion models excel at video generation but are generally limited to short temporal durations. Our theoretical analysis indicates that the autoregressive modeling typically suffers from temporal drift caused by error accumulation and hinders parallelization in long video synthesis. To address these limitations, we propose a novel planning-then-populating framework centered on Macro-from-Micro Planning (MMPL) for long video generation. MMPL sketches a global storyline for the entire video through two hierarchical stages: Micro Planning and Macro Planning. Specifically, Micro Planning predicts a sparse set of future keyframes within each short video segment, offering motion and appearance priors to guide high-quality video segment generation. Macro Planning extends the in-segment keyframes planning across the entire video through an autoregressive chain of micro plans, ensuring long-term consistency across video segments. Subsequently, MMPL-based Content Populating generates all intermediate frames in parallel across segments, enabling efficient parallelization of autoregressive generation. The parallelization is further optimized by Adaptive Workload Scheduling for balanced GPU execution and accelerated autoregressive video generation. Extensive experiments confirm that our method outperforms existing long video generation models in quality and stability. Generated videos and comparison results are in our project page.

Bingyu Li, Da Zhang, Zhiyuan Zhao et al.

Open-vocabulary segmentation aims to identify and segment specific regions and objects based on text-based descriptions. A common solution is to leverage powerful vision-language models (VLMs), such as CLIP, to bridge the gap between vision and text information. However, VLMs are typically pretrained for image-level vision-text alignment, focusing on global semantic features. In contrast, segmentation tasks require fine-grained pixel-level alignment and detailed category boundary information, which VLMs alone cannot provide. As a result, information extracted directly from VLMs can't meet the requirements of segmentation tasks. To address this limitation, we propose FGAseg, a model designed for fine-grained pixel-text alignment and category boundary supplementation. The core of FGAseg is a Pixel-Level Alignment module that employs a cross-modal attention mechanism and a text-pixel alignment loss to refine the coarse-grained alignment from CLIP, achieving finer-grained pixel-text semantic alignment. Additionally, to enrich category boundary information, we introduce the alignment matrices as optimizable pseudo-masks during forward propagation and propose Category Information Supplementation module. These pseudo-masks, derived from cosine and convolutional similarity, provide essential global and local boundary information between different categories. By combining these two strategies, FGAseg effectively enhances pixel-level alignment and category boundary information, addressing key challenges in open-vocabulary segmentation. Extensive experiments demonstrate that FGAseg outperforms existing methods on open-vocabulary semantic segmentation benchmarks.

Bingyu Li, Feiyu Wang, Da Zhang et al.

Most existing underwater instance segmentation approaches are constrained by close-vocabulary prediction, limiting their ability to recognize novel marine categories. To support evaluation, we introduce \textbf{MARIS} (\underline{Mar}ine Open-Vocabulary \underline{I}nstance \underline{S}egmentation), the first large-scale fine-grained benchmark for underwater Open-Vocabulary (OV) segmentation, featuring a limited set of seen categories and diverse unseen categories. Although OV segmentation has shown promise on natural images, our analysis reveals that transfer to underwater scenes suffers from severe visual degradation (e.g., color attenuation) and semantic misalignment caused by lack underwater class definitions. To address these issues, we propose a unified framework with two complementary components. The Geometric Prior Enhancement Module (\textbf{GPEM}) leverages stable part-level and structural cues to maintain object consistency under degraded visual conditions. The Semantic Alignment Injection Mechanism (\textbf{SAIM}) enriches language embeddings with domain-specific priors, mitigating semantic ambiguity and improving recognition of unseen categories. Experiments show that our framework consistently outperforms existing OV baselines both In-Domain and Cross-Domain setting on MARIS, establishing a strong foundation for future underwater perception research.