Xiangdong Zhang

Xiangdong Zhang, Shaofeng Zhang, Junchi Yan

Masked autoencoder has been widely explored in point cloud self-supervised learning, whereby the point cloud is generally divided into visible and masked parts. These methods typically include an encoder accepting visible patches (normalized) and corresponding patch centers (position) as input, with the decoder accepting the output of the encoder and the centers (position) of the masked parts to reconstruct each point in the masked patches. Then, the pre-trained encoders are used for downstream tasks. In this paper, we show a motivating empirical result that when directly feeding the centers of masked patches to the decoder without information from the encoder, it still reconstructs well. In other words, the centers of patches are important and the reconstruction objective does not necessarily rely on representations of the encoder, thus preventing the encoder from learning semantic representations. Based on this key observation, we propose a simple yet effective method, i.e., learning to Predict Centers for Point Masked AutoEncoders (PCP-MAE) which guides the model to learn to predict the significant centers and use the predicted centers to replace the directly provided centers. Specifically, we propose a Predicting Center Module (PCM) that shares parameters with the original encoder with extra cross-attention to predict centers. Our method is of high pre-training efficiency compared to other alternatives and achieves great improvement over Point-MAE, particularly surpassing it by 5.50% on OBJ-BG, 6.03% on OBJ-ONLY, and 5.17% on PB-T50-RS for 3D object classification on the ScanObjectNN dataset. The code is available at https://github.com/aHapBean/PCP-MAE.

Xiangdong Zhang, Debing Zhang, Shaofeng Zhang et al.

Standard next-token prediction (NTP) supervises language models solely through discrete labels in the output logit space. We argue that this sparse one-hot supervision leaves the latent representation space under-constrained, allowing hidden states to drift into degenerate and anisotropic configurations that can limit generalization. To address this issue, we propose Next Implicit Token Prediction (NITP), which augments discrete prediction with dense continuous supervision directly in the representation space. NITP trains the model to predict the implicit semantic content of the next token, using shallow-layer representations from the same model as stable self-supervised targets. We provide theoretical analysis showing that NITP regularizes the optimization landscape by mitigating under-constrained degrees of freedom and encouraging a compact, structured representation geometry. Empirically, across dense and MoE models ranging from 0.5B to 9B parameters, NITP consistently improves downstream performance with negligible computational overhead. On a 9B MoE model, NITP achieves a 5.7% absolute improvement on MMLU-Pro, along with gains of 6.4% on C3 and 4.3% on CommonsenseQA, with approximately 2% additional training FLOPs and no additional inference cost. Our implementation is available at https://github.com/aHapBean/NITP.

Jingtao Ye, Kexin Zhang, Xunchi Ma et al.

The rapid movements and agile maneuvers of unmanned aerial vehicles (UAVs) induce significant observational challenges for multi-object tracking (MOT). However, existing UAV-perspective MOT benchmarks often lack these complexities, featuring predominantly predictable camera dynamics and linear motion patterns. To address this gap, we introduce DynUAV, a new benchmark for dynamic UAV-perspective MOT, characterized by intense ego-motion and the resulting complex apparent trajectories. The benchmark comprises 42 video sequences with over 1.7 million bounding box annotations, covering vehicles, pedestrians, and specialized industrial categories such as excavators, bulldozers and cranes. Compared to existing benchmarks, DynUAV introduces substantial challenges arising from ego-motion, including drastic scale changes and viewpoint changes, as well as motion blur. Comprehensive evaluations of state-of-the-art trackers on DynUAV reveal their limitations, particularly in managing the intertwined challenges of detection and association under such dynamic conditions, thereby establishing DynUAV as a rigorous benchmark. We anticipate that DynUAV will serve as a demanding testbed to spur progress in real-world UAV-perspective MOT, and we will make all resources available at link.

Shaofeng Zhang, Xuanqi Chen, Xiangdong Zhang et al.

Most existing self-supervised learning (SSL) approaches for 3D point clouds are dominated by generative methods based on Masked Autoencoders (MAE). However, these generative methods have been proven to struggle to capture high-level discriminative features effectively, leading to poor performance on linear probing and other downstream tasks. In contrast, contrastive methods excel in discriminative feature representation and generalization ability on image data. Despite this, contrastive learning (CL) in 3D data remains scarce. Besides, simply applying CL methods designed for 2D data to 3D fails to effectively learn 3D local details. To address these challenges, we propose a novel Dual-Branch \textbf{C}enter-\textbf{S}urrounding \textbf{Con}trast (CSCon) framework. Specifically, we apply masking to the center and surrounding parts separately, constructing dual-branch inputs with center-biased and surrounding-biased representations to better capture rich geometric information. Meanwhile, we introduce a patch-level contrastive loss to further enhance both high-level information and local sensitivity. Under the FULL and ALL protocols, CSCon achieves performance comparable to generative methods; under the MLP-LINEAR, MLP-3, and ONLY-NEW protocols, our method attains state-of-the-art results, even surpassing cross-modal approaches. In particular, under the MLP-LINEAR protocol, our method outperforms the baseline (Point-MAE) by \textbf{7.9\%}, \textbf{6.7\%}, and \textbf{10.3\%} on the three variants of ScanObjectNN, respectively. The code will be made publicly available.

Shaobo Wang, Xiangdong Zhang, Dongrui Liu et al.

Batch Normalization (BN) has become an essential technique in contemporary neural network design, enhancing training stability. Specifically, BN employs centering and scaling operations to standardize features along the batch dimension and uses an affine transformation to recover features. Although standard BN has shown its capability to improve deep neural network training and convergence, it still exhibits inherent limitations in certain cases. Current enhancements to BN typically address only isolated aspects of its mechanism. In this work, we critically examine BN from a feature perspective, identifying feature condensation during BN as a detrimental factor to test performance. To tackle this problem, we propose a two-stage unified framework called Unified Batch Normalization (UBN). In the first stage, we employ a straightforward feature condensation threshold to mitigate condensation effects, thereby preventing improper updates of statistical norms. In the second stage, we unify various normalization variants to boost each component of BN. Our experimental results reveal that UBN significantly enhances performance across different visual backbones and different vision tasks, and notably expedites network training convergence, particularly in early training stages. Notably, our method improved about 3% in accuracy on ImageNet classification and 4% in mean average precision on both Object Detection and Instance Segmentation on COCO dataset, showing the effectiveness of our approach in real-world scenarios.

Xiangdong Zhang, Shaofeng Zhang, Junchi Yan

Point cloud learning, especially in a self-supervised way without manual labels, has gained growing attention in both vision and learning communities due to its potential utility in a wide range of applications. Most existing generative approaches for point cloud self-supervised learning focus on recovering masked points from visible ones within a single view. Recognizing that a two-view pre-training paradigm inherently introduces greater diversity and variance, it may thus enable more challenging and informative pre-training. Inspired by this, we explore the potential of two-view learning in this domain. In this paper, we propose Point-PQAE, a cross-reconstruction generative paradigm that first generates two decoupled point clouds/views and then reconstructs one from the other. To achieve this goal, we develop a crop mechanism for point cloud view generation for the first time and further propose a novel positional encoding to represent the 3D relative position between the two decoupled views. The cross-reconstruction significantly increases the difficulty of pre-training compared to self-reconstruction, which enables our method to surpass previous single-modal self-reconstruction methods in 3D self-supervised learning. Specifically, it outperforms the self-reconstruction baseline (Point-MAE) by 6.5%, 7.0%, and 6.7% in three variants of ScanObjectNN with the Mlp-Linear evaluation protocol. The code is available at https://github.com/aHapBean/Point-PQAE.

Yanbin Yin, Kun Zhou, Zhen Wang et al.

The recent explosion of large language models (LLMs), each with its own general or specialized strengths, makes scalable, reliable benchmarking more urgent than ever. Standard practices nowadays face fundamental trade-offs: closed-ended question-based benchmarks (eg MMLU) struggle with saturation as newer models emerge, while crowd-sourced leaderboards (eg Chatbot Arena) rely on costly and slow human judges. Recently, automated methods (eg LLM-as-a-judge) shed light on the scalability, but risk bias by relying on one or a few "authority" models. To tackle these issues, we propose Decentralized Arena (dearena), a fully automated framework leveraging collective intelligence from all LLMs to evaluate each other. It mitigates single-model judge bias by democratic, pairwise evaluation, and remains efficient at scale through two key components: (1) a coarse-to-fine ranking algorithm for fast incremental insertion of new models with sub-quadratic complexity, and (2) an automatic question selection strategy for the construction of new evaluation dimensions. Across extensive experiments across 66 LLMs, dearena attains up to 97% correlation with human judgements, while significantly reducing the cost. Our code and data will be publicly released on https://github.com/maitrix-org/de-arena.

Mingtao Feng, Zhen Li, Qi Li et al.

3D object grounding aims to locate the most relevant target object in a raw point cloud scene based on a free-form language description. Understanding complex and diverse descriptions, and lifting them directly to a point cloud is a new and challenging topic due to the irregular and sparse nature of point clouds. There are three main challenges in 3D object grounding: to find the main focus in the complex and diverse description; to understand the point cloud scene; and to locate the target object. In this paper, we address all three challenges. Firstly, we propose a language scene graph module to capture the rich structure and long-distance phrase correlations. Secondly, we introduce a multi-level 3D proposal relation graph module to extract the object-object and object-scene co-occurrence relationships, and strengthen the visual features of the initial proposals. Lastly, we develop a description guided 3D visual graph module to encode global contexts of phrases and proposals by a nodes matching strategy. Extensive experiments on challenging benchmark datasets (ScanRefer and Nr3D) show that our algorithm outperforms existing state-of-the-art. Our code is available at https://github.com/PNXD/FFL-3DOG.

Xiangdong Zhang, Jiaqi Liao, Shaofeng Zhang et al.

Recent advancements in text-to-video (T2V) diffusion models have enabled high-fidelity and realistic video synthesis. However, current T2V models often struggle to generate physically plausible content due to their limited inherent ability to accurately understand physics. We found that while the representations within T2V models possess some capacity for physics understanding, they lag significantly behind those from recent video self-supervised learning methods. To this end, we propose a novel framework called VideoREPA, which distills physics understanding capability from video understanding foundation models into T2V models by aligning token-level relations. This closes the physics understanding gap and enable more physics-plausible generation. Specifically, we introduce the Token Relation Distillation (TRD) loss, leveraging spatio-temporal alignment to provide soft guidance suitable for finetuning powerful pre-trained T2V models, a critical departure from prior representation alignment (REPA) methods. To our knowledge, VideoREPA is the first REPA method designed for finetuning T2V models and specifically for injecting physical knowledge. Empirical evaluations show that VideoREPA substantially enhances the physics commonsense of baseline method, CogVideoX, achieving significant improvement on relevant benchmarks and demonstrating a strong capacity for generating videos consistent with intuitive physics. More video results are available at https://videorepa.github.io/.

Xiaohan Qin, Xiaoxing Wang, Ning Liao et al.

Data quality plays a critical role in enhancing supervised fine-tuning (SFT) for large language models (LLMs), and token-level data selection has emerged as a promising direction for its fine-grained nature. Despite their strong empirical performance, existing token-level selection methods share two key limitations: (1) requiring training or accessing an additional reference model, and (2) relying solely on loss information for token selection, which cannot well preserve semantically important tokens that are not favored by loss-based metrics. To address these challenges, we propose ssToken, a Self-modulated and Semantic-aware Token Selection approach. ssToken leverages readily accessible history models to compute the per-token loss difference with the current model, which serves as a self-modulated signal that enables the model to adaptively select tokens along its optimization trajectory, rather than relying on excess loss from an offline-trained reference model as in prior works. We further introduce a semantic-aware, attention-based token importance estimation metric, orthogonal to loss-based selection and providing complementary semantic information for more effective filtering. Extensive experiments across different model families and scales demonstrate that both self-modulated selection and semantic-aware selection alone outperform full-data fine-tuning, while their integration--ssToken--achieves synergistic gains and further surpasses prior token-level selection methods, delivering performance improvements while maintaining training efficiency.

Jing Qiu, Jiancheng Huang, Xiangdong Zhang et al.

Physics-informed deep learning has been developed as a novel paradigm for learning physical dynamics recently. While general physics-informed deep learning methods have shown early promise in learning fluid dynamics, they are difficult to generalize in arbitrary time instants in real-world scenario, where the fluid motion can be considered as a time-variant trajectory involved large-scale particles. Inspired by the advantage of diffusion model in learning the distribution of data, we first propose Pi-fusion, a physics-informed diffusion model for predicting the temporal evolution of velocity and pressure field in fluid dynamics. Physics-informed guidance sampling is proposed in the inference procedure of Pi-fusion to improve the accuracy and interpretability of learning fluid dynamics. Furthermore, we introduce a training strategy based on reciprocal learning to learn the quasiperiodical pattern of fluid motion and thus improve the generalizability of the model. The proposed approach are then evaluated on both synthetic and real-world dataset, by comparing it with state-of-the-art physics-informed deep learning methods. Experimental results show that the proposed approach significantly outperforms existing methods for predicting temporal evolution of velocity and pressure field, confirming its strong generalization by drawing probabilistic inference of forward process and physics-informed guidance sampling. The proposed Pi-fusion can also be generalized in learning other physical dynamics governed by partial differential equations.

Quanxue Gao, Wei Xia, Xinbo Gao et al.

Despite the impressive clustering performance and efficiency in characterizing both the relationship between data and cluster structure, existing graph-based multi-view clustering methods still have the following drawbacks. They suffer from the expensive time burden due to both the construction of graphs and eigen-decomposition of Laplacian matrix, and fail to explore the cluster structure of large-scale data. Moreover, they require a post-processing to get the final clustering, resulting in suboptimal performance. Furthermore, rank of the learned view-consensus graph cannot approximate the target rank. In this paper, drawing the inspiration from the bipartite graph, we propose an effective and efficient graph learning model for multi-view clustering. Specifically, our method exploits the view-similar between graphs of different views by the minimization of tensor Schatten p-norm, which well characterizes both the spatial structure and complementary information embedded in graphs of different views. We learn view-consensus graph with adaptively weighted strategy and connectivity constraint such that the connected components indicates clusters directly. Our proposed algorithm is time-economical and obtains the stable results and scales well with the data size. Extensive experimental results indicate that our method is superior to state-of-the-art methods.

Xiangdong Zhang, Tengjun Wang, Yun Yang

Because hyperspectral remote sensing images contain a lot of redundant information and the data structure is highly non-linear, leading to low classification accuracy of traditional machine learning methods. The latest research shows that hyperspectral image classification based on deep convolutional neural network has high accuracy. However, when a small amount of data is used for training, the classification accuracy of deep learning methods is greatly reduced. In order to solve the problem of low classification accuracy of existing algorithms on small samples of hyperspectral images, a multi-scale residual network is proposed. The multi-scale extraction and fusion of spatial and spectral features is realized by adding a branch structure into the residual block and using convolution kernels of different sizes in the branch. The spatial and spectral information contained in hyperspectral images are fully utilized to improve the classification accuracy. In addition, in order to improve the speed and prevent overfitting, the model uses dynamic learning rate, BN and Dropout strategies. The experimental results show that the overall classification accuracy of this method is 99.07% and 99.96% respectively in the data set of Indian Pines and Pavia University, which is better than other algorithms.