Shaofeng Zhang

Qiang Zhou, Chaohui Yu, Shaofeng Zhang et al.

In this work, we investigate extending the comprehension of Multi-modal Large Language Models (MLLMs) to regional objects. To this end, we propose to extract features corresponding to regional objects as soft prompts for LLM, which provides a straightforward and scalable approach and eliminates the need for LLM fine-tuning. To effectively extract regional features from regular image features and irregular point cloud features, we present a novel and unified position-assisted feature extraction module. Furthermore, training an MLLM from scratch is highly time-consuming. Thus, we propose incrementally extending existing pre-trained MLLMs to comprehend more modalities and the regional objects of those modalities. Specifically, we freeze the Q-Former from BLIP-2, an impressive MLLM, and optimize the modality-specific Lora parameters in Q-Former and LLM for each newly introduced modality. The freezing of the Q-Former eliminates the need for extensive pre-training on massive image-text data. The freezed Q-Former pre-trained from massive image-text data is also beneficial for the pre-training on image-region-text data. We name our framework RegionBLIP. We pre-train RegionBLIP on image-region-text, point-cloud-text, and point-cloud-region-text data. Experimental results verify that \Ours{} can preserve the image comprehension capability of BILP-2 and further gain a comprehension of the newly introduced point cloud modality and regional objects. The Data, Code, and Pre-trained models will be available at https://github.com/mightyzau/RegionBLIP.

Xiaolin Liu, Yilun Zhu, Xiangyu Zhao et al.

Video multimodal large language models (MLLMs) have made rapid progress on general and long-form video understanding, yet their ability to preserve brief answer-critical visual evidence remains underexplored. Many practical questions are determined by momentary visual events: localized actions or state transitions that may last only a few frames. Such evidence can be skipped by sparse frame sampling, suppressed by visual-token compression, or diluted by coarse temporal aggregation, causing failures that language-side reasoning cannot reliably recover. We introduce Moment-Video, a benchmark for diagnosing the temporal fidelity of video MLLMs through momentary visual event understanding. Each question is grounded in a localized, visually observable, and sampling-sensitive event, requiring models to notice, count, describe, or reason about transient evidence rather than rely on persistent objects, global scene context, or language priors. Moment-Video contains 1,000 human-verified video-QA pairs across 7 domains and 25 fine-grained subcategories, covering four task types: Temporal Occurrence, Temporal Counting, Action Description, and Temporal Reasoning. We evaluate 33 proprietary and open-source MLLMs on Moment-Video. The best-performing model, Seed-2.0-Pro, achieves only 39.6% overall accuracy, while most open-source models remain below 25%, revealing a substantial gap in momentary visual event understanding. Diagnostic analyses show that denser frame sampling improves some models but does not eliminate the bottleneck, and longer videos introduce stronger temporal-localization challenges. These findings suggest that current video MLLMs still lack temporally faithful representations for capturing, preserving, and using brief but decisive visual evidence.

Shaofeng Zhang, Feng Zhu, Junchi Yan et al.

Scalability is an important consideration for deep graph neural networks. Inspired by the conventional pooling layers in CNNs, many recent graph learning approaches have introduced the pooling strategy to reduce the size of graphs for learning, such that the scalability and efficiency can be improved. However, these pooling-based methods are mainly tailored to a single graph-level task and pay more attention to local information, limiting their performance in multi-task settings which often require task-specific global information. In this paper, departure from these pooling-based efforts, we design a new approach called DOTIN (\underline{D}r\underline{o}pping \underline{T}ask-\underline{I}rrelevant \underline{N}odes) to reduce the size of graphs. Specifically, by introducing $K$ learnable virtual nodes to represent the graph embeddings targeted to $K$ different graph-level tasks, respectively, up to 90\% raw nodes with low attentiveness with an attention model -- a transformer in this paper, can be adaptively dropped without notable performance decreasing. Achieving almost the same accuracy, our method speeds up GAT by about 50\% on graph-level tasks including graph classification and graph edit distance (GED) with about 60\% less memory, on D\&D dataset. Code will be made publicly available in https://github.com/Sherrylone/DOTIN.

Junkun Yuan, Xinyu Zhang, Hao Zhou et al. · tencent-ai

Model pre-training is essential in human-centric perception. In this paper, we first introduce masked image modeling (MIM) as a pre-training approach for this task. Upon revisiting the MIM training strategy, we reveal that human structure priors offer significant potential. Motivated by this insight, we further incorporate an intuitive human structure prior - human parts - into pre-training. Specifically, we employ this prior to guide the mask sampling process. Image patches, corresponding to human part regions, have high priority to be masked out. This encourages the model to concentrate more on body structure information during pre-training, yielding substantial benefits across a range of human-centric perception tasks. To further capture human characteristics, we propose a structure-invariant alignment loss that enforces different masked views, guided by the human part prior, to be closely aligned for the same image. We term the entire method as HAP. HAP simply uses a plain ViT as the encoder yet establishes new state-of-the-art performance on 11 human-centric benchmarks, and on-par result on one dataset. For example, HAP achieves 78.1% mAP on MSMT17 for person re-identification, 86.54% mA on PA-100K for pedestrian attribute recognition, 78.2% AP on MS COCO for 2D pose estimation, and 56.0 PA-MPJPE on 3DPW for 3D pose and shape estimation.

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.

Shaofeng Zhang, Feng Zhu, Rui Zhao et al.

We propose ADCLR: A ccurate and D ense Contrastive Representation Learning, a novel self-supervised learning framework for learning accurate and dense vision representation. To extract spatial-sensitive information, ADCLR introduces query patches for contrasting in addition with global contrasting. Compared with previous dense contrasting methods, ADCLR mainly enjoys three merits: i) achieving both global-discriminative and spatial-sensitive representation, ii) model-efficient (no extra parameters in addition to the global contrasting baseline), and iii) correspondence-free and thus simpler to implement. Our approach achieves new state-of-the-art performance for contrastive methods. On classification tasks, for ViT-S, ADCLR achieves 77.5% top-1 accuracy on ImageNet with linear probing, outperforming our baseline (DINO) without our devised techniques as plug-in, by 0.5%. For ViT-B, ADCLR achieves 79.8%, 84.0% accuracy on ImageNet by linear probing and finetune, outperforming iBOT by 0.3%, 0.2% accuracy. For dense tasks, on MS-COCO, ADCLR achieves significant improvements of 44.3% AP on object detection, 39.7% AP on instance segmentation, outperforming previous SOTA method SelfPatch by 2.2% and 1.2%, respectively. On ADE20K, ADCLR outperforms SelfPatch by 1.0% mIoU, 1.2% mAcc on the segme

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.

Tencent Hunyuan Team, Ao Liu, Botong Zhou et al. · tencent-ai

As Large Language Models (LLMs) rapidly advance, we introduce Hunyuan-TurboS, a novel large hybrid Transformer-Mamba Mixture of Experts (MoE) model. It synergistically combines Mamba's long-sequence processing efficiency with Transformer's superior contextual understanding. Hunyuan-TurboS features an adaptive long-short chain-of-thought (CoT) mechanism, dynamically switching between rapid responses for simple queries and deep "thinking" modes for complex problems, optimizing computational resources. Architecturally, this 56B activated (560B total) parameter model employs 128 layers (Mamba2, Attention, FFN) with an innovative AMF/MF block pattern. Faster Mamba2 ensures linear complexity, Grouped-Query Attention minimizes KV cache, and FFNs use an MoE structure. Pre-trained on 16T high-quality tokens, it supports a 256K context length and is the first industry-deployed large-scale Mamba model. Our comprehensive post-training strategy enhances capabilities via Supervised Fine-Tuning (3M instructions), a novel Adaptive Long-short CoT Fusion method, Multi-round Deliberation Learning for iterative improvement, and a two-stage Large-scale Reinforcement Learning process targeting STEM and general instruction-following. Evaluations show strong performance: overall top 7 rank on LMSYS Chatbot Arena with a score of 1356, outperforming leading models like Gemini-2.0-Flash-001 (1352) and o4-mini-2025-04-16 (1345). TurboS also achieves an average of 77.9% across 23 automated benchmarks. Hunyuan-TurboS balances high performance and efficiency, offering substantial capabilities at lower inference costs than many reasoning models, establishing a new paradigm for efficient large-scale pre-trained models.

Mingxin Liu, Ziqian Fan, Zhaokai Wang et al.

Unified multimodal models target joint understanding, reasoning, and generation, but current image editing benchmarks are largely confined to natural images and shallow commonsense reasoning, offering limited assessment of this capability under structured, domain-specific constraints. In this work, we introduce GRADE, the first benchmark to assess discipline-informed knowledge and reasoning in image editing. GRADE comprises 520 carefully curated samples across 10 academic domains, spanning from natural science to social science. To support rigorous evaluation, we propose a multi-dimensional evaluation protocol that jointly assesses Discipline Reasoning, Visual Consistency, and Logical Readability. Extensive experiments on 20 state-of-the-art open-source and closed-source models reveal substantial limitations in current models under implicit, knowledge-intensive editing settings, leading to large performance gaps. Beyond quantitative scores, we conduct rigorous analyses and ablations to expose model shortcomings and identify the constraints within disciplinary editing. Together, GRADE pinpoints key directions for the future development of unified multimodal models, advancing the research on discipline-informed image editing and reasoning. Our benchmark and evaluation code are publicly released.

Jinpei Guo, Shaofeng Zhang, Runzhong Wang et al. · pku

Vision transformers (ViTs) have recently been used for visual matching beyond object detection and segmentation. However, the original grid dividing strategy of ViTs neglects the spatial information of the keypoints, limiting the sensitivity to local information. Therefore, we propose QueryTrans (Query Transformer), which adopts a cross-attention module and keypoints-based center crop strategy for better spatial information extraction. We further integrate the graph attention module and devise a transformer-based graph matching approach GMTR (Graph Matching TRansformers) whereby the combinatorial nature of GM is addressed by a graph transformer neural GM solver. On standard GM benchmarks, GMTR shows competitive performance against the SOTA frameworks. Specifically, on Pascal VOC, GMTR achieves $\mathbf{83.6\%}$ accuracy, $\mathbf{0.9\%}$ higher than the SOTA framework. On Spair-71k, GMTR shows great potential and outperforms most of the previous works. Meanwhile, on Pascal VOC, QueryTrans improves the accuracy of NGMv2 from $80.1\%$ to $\mathbf{83.3\%}$, and BBGM from $79.0\%$ to $\mathbf{84.5\%}$. On Spair-71k, QueryTrans improves NGMv2 from $80.6\%$ to $\mathbf{82.5\%}$, and BBGM from $82.1\%$ to $\mathbf{83.9\%}$. Source code will be made publicly available.

Xiaosong Jia, Chenhe Zhang, Yule Jiang et al.

Existing autonomous driving systems rely on onboard sensors (cameras, LiDAR, IMU, etc) for environmental perception. However, this paradigm is limited by the drive-time perception horizon and often fails under limited view scope, occlusion or extreme conditions such as darkness and rain. In contrast, human drivers are able to recall road structure even under poor visibility. To endow models with this ``recall" ability, we propose the spatial retrieval paradigm, introducing offline retrieved geographic images as an additional input. These images are easy to obtain from offline caches (e.g, Google Maps or stored autonomous driving datasets) without requiring additional sensors, making it a plug-and-play extension for existing AD tasks. For experiments, we first extend the nuScenes dataset with geographic images retrieved via Google Maps APIs and align the new data with ego-vehicle trajectories. We establish baselines across five core autonomous driving tasks: object detection, online mapping, occupancy prediction, end-to-end planning, and generative world modeling. Extensive experiments show that the extended modality could enhance the performance of certain tasks. We will open-source dataset curation code, data, and benchmarks for further study of this new autonomous driving paradigm.

Hengrui Zhang, Qitian Wu, Yu Wang et al.

Contrastive learning methods based on InfoNCE loss are popular in node representation learning tasks on graph-structured data. However, its reliance on data augmentation and its quadratic computational complexity might lead to inconsistency and inefficiency problems. To mitigate these limitations, in this paper, we introduce a simple yet effective contrastive model named Localized Graph Contrastive Learning (Local-GCL in short). Local-GCL consists of two key designs: 1) We fabricate the positive examples for each node directly using its first-order neighbors, which frees our method from the reliance on carefully-designed graph augmentations; 2) To improve the efficiency of contrastive learning on graphs, we devise a kernelized contrastive loss, which could be approximately computed in linear time and space complexity with respect to the graph size. We provide theoretical analysis to justify the effectiveness and rationality of the proposed methods. Experiments on various datasets with different scales and properties demonstrate that in spite of its simplicity, Local-GCL achieves quite competitive performance in self-supervised node representation learning tasks on graphs with various scales and properties.

Qihong Tang, Changhan Liu, Shaofeng Zhang et al.

Identifying potential objects is critical for object recognition and analysis across various computer vision applications. Existing methods typically localize potential objects by relying on exemplar images, predefined categories, or textual descriptions. However, their reliance on image and text prompts often limits flexibility, restricting adaptability in real-world scenarios. In this paper, we introduce a novel Prompt-Free Universal Region Proposal Network (PF-RPN), which identifies potential objects without relying on external prompts. First, the Sparse Image-Aware Adapter (SIA) module performs initial localization of potential objects using a learnable query embedding dynamically updated with visual features. Next, the Cascade Self-Prompt (CSP) module identifies the remaining potential objects by leveraging the self-prompted learnable embedding, autonomously aggregating informative visual features in a cascading manner. Finally, the Centerness-Guided Query Selection (CG-QS) module facilitates the selection of high-quality query embeddings using a centerness scoring network. Our method can be optimized with limited data (e.g., 5% of MS COCO data) and applied directly to various object detection application domains for identifying potential objects without fine-tuning, such as underwater object detection, industrial defect detection, and remote sensing image object detection. Experimental results across 19 datasets validate the effectiveness of our method. Code is available at https://github.com/tangqh03/PF-RPN.

Shaofeng Zhang, Shengcai Liu, Ning Lu et al.

Combinatorial optimization problems are widely encountered in real-world applications. A critical research challenge lies in designing high-quality heuristic algorithms that efficiently approximate optimal solutions within a reasonable time. In recent years, many works have explored integrating Large Language Models (LLMs) with Evolutionary Algorithms to automate heuristic algorithm design through prompt engineering. However, these approaches generally adopt a problem-specific paradigm, applying a single algorithm across all problem instances, failing to account for the heterogeneity across instances. In this paper, we propose InstSpecHH, a novel framework that introduces the concept of instance-specific heuristic generation. InstSpecHH partitions the overall problem class into sub-classes based on instance features and performs differentiated, automated heuristic design for each problem subclass. By tailoring heuristics to the unique features of different sub-classes, InstSpecHH achieves better performance at the problem class level while avoiding redundant heuristic generation for similar instances, thus reducing computational overhead. This approach effectively balances the trade-off between the cost of automatic heuristic design and the quality of the obtained solutions. To evaluate the performance of InstSpecHH, we conduct comprehensive experiments on 4,500 subclasses of the Online Bin Packing Problem (OBPP) and 365 subclasses of the Capacitated Vehicle Routing Problem (CVRP). Experimental results show that InstSpecHH demonstrates strong intra-subclass and inter-subclass generalization capabilities. Compared to previous problem-specific methods, InstSpecHH reduces the average optimality gap by 6.06\% for OBPP and 0.66\% for CVRP. These results highlight the potential of instance-aware automatic heuristic design to further enhance solution quality.

Shanwen Tan, Hao Li, Jingtao Zhang et al.

Streaming long-video generation faces a central challenge in continuous semantic switching, requiring adaptive memory to preserve coherent visual evolution. Current approaches rely on cache rebuilding at prompt boundaries or fixed memory budgets, but they introduce redundant computation and limit flexible semantic adaptation. This limitation arises from a mismatch between cached video history and prompt updates, as memory preserves visual continuity while prompt switches demand rapid semantic adaptation. Motivated by this observation, we present SWIFT, Semantic Windowing and Injection for Flexible Transitions, a training-free framework for multi-prompt long-video generation that enables efficient semantic switching while preserving temporal coherence in causal video diffusion models. SWIFT introduces a lightweight Semantic Injection Cache that augments cached video memory rather than reconstructing it from scratch at every prompt boundary. To avoid uniformly perturbing all attention channels, we further perform head-wise semantic injection, so that each attention head receives a prompt update proportional to its alignment with the current video state. In addition, we introduce an Adaptive Dynamic Window that allocates temporal memory according to prompt phase, using larger local context near switching boundaries and smaller windows during stable segments to reduce average inference cost. To preserve long-range semantic consistency under compressed local attention, we further maintain segment-level semantic anchors that summarize prompt-conditioned video history and reintroduce it as compact memory tokens. Compared with current state-of-the-art methods, SWIFT preserves generation quality while achieving 22.6 FPS on a single H100 GPU, establishing a substantially more efficient solution for multi-prompt long-video generation. Our code is available at https://github.com/ShanwenTan/SWIFT.

Shaofeng Zhang, Xuanqi Chen, Ning Liao et al.

The dominance of denoising generative models (e.g., diffusion, flow-matching) in visual synthesis is tempered by their substantial training costs and inefficiencies in representation learning. While injecting discriminative representations via auxiliary alignment has proven effective, this approach still faces key limitations: the reliance on external, pre-trained encoders introduces overhead and domain shift. A dispersed-based strategy that encourages strong separation among in-batch latent representations alleviates this specific dependency. To assess the effect of the number of negative samples in generative modeling, we propose {\mname}, a plug-and-play training framework that requires no external encoders. Our method integrates a memory bank mechanism that maintains a large, dynamically updated queue of negative samples across training iterations. This decouples the number of negatives from the mini-batch size, providing abundant and high-quality negatives for a contrastive objective without a multiplicative increase in computational cost. A low-dimensional projection head is used to further minimize memory and bandwidth overhead. {\mname} offers three principal advantages: (1) it is self-contained, eliminating dependency on pretrained vision foundation models and their associated forward-pass overhead; (2) it introduces no additional parameters or computational cost during inference; and (3) it enables substantially faster convergence, achieving superior generative quality more efficiently. On ImageNet-256, {\mname} achieves a state-of-the-art FID of \textbf{2.40} within 400k steps, significantly outperforming comparable methods.

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.

Ning Liao, Shaofeng Zhang, Renqiu Xia et al.

There is an emerging line of research on multimodal instruction tuning, and a line of benchmarks has been proposed for evaluating these models recently. Instead of evaluating the models directly, in this paper, we try to evaluate the Vision-Language Instruction-Tuning (VLIT) datasets. Also, we seek the way of building a dataset for developing an all-powerful VLIT model, which we believe could also be of utility for establishing a grounded protocol for benchmarking VLIT models. For effective evaluation of VLIT datasets that remains an open question, we propose a tune-cross-evaluation paradigm: tuning on one dataset and evaluating on the others in turn. For each single tune-evaluation experiment set, we define the Meta Quality (MQ) as the mean score obtained by a set of caption metrics including BLEU, METEOR, and ROUGE-L to quantify the quality of a certain dataset or a sample. On this basis, to evaluate the comprehensiveness of a dataset, we develop the Dataset Quality (DQ) covering all tune-evaluation sets. To lay the foundation for building a comprehensive dataset and developing an all-powerful model for practical applications, we define the Sample Quality (SQ) to quantify the all-sided quality of each sample. Extensive experiments validate the rationality of the proposed evaluation paradigm. Based on the holistic evaluation, we build a new dataset, REVO-LION (REfining VisiOn-Language InstructiOn tuNing), by collecting samples with higher SQ from each dataset. Remarkably, even with only half of the complete data, the model trained on REVO-LION can achieve the performance comparable to simply adding all VLIT datasets up. Furthermore, REVO-LION not only facilitates the development of a powerful model but also incorporates an evaluation set, which is designed to serve as a convenient benchmark for future research in the field.

Yi Yu, Botao Ren, Peiyuan Zhang et al.

With the rapidly increasing demand for oriented object detection (OOD), recent research involving weakly-supervised detectors for learning OOD from point annotations has gained great attention. In this paper, we rethink this challenging task setting with the layout among instances and present Point2RBox-v2. At the core are three principles: 1) Gaussian overlap loss. It learns an upper bound for each instance by treating objects as 2D Gaussian distributions and minimizing their overlap. 2) Voronoi watershed loss. It learns a lower bound for each instance through watershed on Voronoi tessellation. 3) Consistency loss. It learns the size/rotation variation between two output sets with respect to an input image and its augmented view. Supplemented by a few devised techniques, e.g. edge loss and copy-paste, the detector is further enhanced. To our best knowledge, Point2RBox-v2 is the first approach to explore the spatial layout among instances for learning point-supervised OOD. Our solution is elegant and lightweight, yet it is expected to give a competitive performance especially in densely packed scenes: 62.61%/86.15%/34.71% on DOTA/HRSC/FAIR1M. Code is available at https://github.com/VisionXLab/point2rbox-v2.

Yihang Wu, Yihang Sun, Shaofeng Zhang et al.

Transformer-based models have advanced feedforward novel view synthesis (NVS). Current architectures such as GS-LRM and LVSM mix semantic information (e.g., RGB) and spatial information (e.g., Plücker rays) into a shared feature space. Since Plücker rays naturally carry lattice-like spatial structure, these designs can make the spatial bias interfere with appearance representation and degrade rendering fidelity. To this end, we propose to decouple the representation of feedforward NVS transformers into separate semantic and spatial tokens. The decoupled design keeps semantic and spatial information explicit in their branches while preserving cross-branch interaction through shared attention routing. Built on this design, we introduce optional categorized supervision and bidirectional modulation: the former provides branch-specific training signals, while the latter improves interaction between the two branches. Notably, the base decoupled design introduces virtually zero additional inference latency due to its architectural design. The proposed designs achieve consistent improvements, demonstrating effectiveness across decoder-only and encoder-decoder feedforward NVS models.

Zhiyuan Zhang, Zhenghao Jin, Yanlun Peng et al.

Robustness is a critical requirement for deploying autonomous driving systems in the real world. Existing robustness benchmarks for autonomous driving have made important progress in studying the effects of image-level corruptions, such as adverse weather or camera degradation, on perception modules and open-loop planning outputs. However, deployment can also involve system-level imperfections, such as inference latency and ego-state estimation errors, which remain less studied in closed-loop E2E-AD evaluation. These imperfections can accumulate through the feedback loop and destabilize control. In this work, we present Bench2Drive-Robust, to our knowledge the first device-centric robustness benchmark for closed-loop end-to-end autonomous driving under realistic deployment perturbations. We systematically evaluate deployment-oriented perturbations arising from three major sources: camera-stream failures (frame drop, partial observation), ego-state estimation errors (GPS noise, and speed or odometry errors), and compute-induced control delay (model inference delay). We evaluate representative end-to-end driving methods and analyze their robustness under different perturbation severities. Our results show that these deployment-related perturbations can substantially degrade closed-loop driving performance, revealing robustness challenges that are not fully captured by conventional image-level corruption evaluations. By establishing a closed-loop evaluation protocol and demonstrating the substantial impact of these deployment-oriented perturbations, Bench2Drive-Robust defines practical robustness problems for end-to-end autonomous driving and encourages further research on deployment-aware robust driving systems.

Yan Li, Ning Liao, Xiangyu Zhao et al.

The development of unified multimodal large language models (MLLMs) is fundamentally challenged by the granularity gap between visual understanding and generation: understanding requires high-level semantic abstractions, while image generation demands fine-grained pixel-level representations. Existing approaches usually enforce the two supervision on the same set of representation or decouple these two supervision on separate feature spaces, leading to interference and inconsistency, respectively. In this work, we propose EvoTok, a unified image tokenizer that reconciles these requirements through a residual evolution process within a shared latent space. Instead of maintaining separate token spaces for pixels and semantics, EvoTok encodes an image into a cascaded sequence of residual tokens via residual vector quantization. This residual sequence forms an evolution trajectory where earlier stages capture low-level details and deeper stages progressively transition toward high-level semantic representations. Despite being trained on a relatively modest dataset of 13M images, far smaller than the billion-scale datasets used by many previous unified tokenizers, EvoTok achieves a strong reconstruction quality of 0.43 rFID on ImageNet-1K at 256x256 resolution. When integrated with a large language model, EvoTok shows promising performance across 7 out of 9 visual understanding benchmarks, and remarkable results on image generation benchmarks such as GenEval and GenAI-Bench. These results demonstrate that modeling visual representations as an evolving trajectory provides an effective and principled solution for unifying visual understanding and generation.

Zhiyuan Wang, Shengcai Liu, Shaofeng Zhang et al.

Despite recent progress in constructing generalizable parallel algorithm portfolios (PAPs), no general-purpose approach is yet available for multi-objective binary optimization problems (MOBOPs). To fill this gap, this paper proposes domain-agnostic co-evolution of parameterized search for multi-objective binary optimization~(DACMO), which features two technical innovations. First, we propose a neural instance representation architecture that decouples domain-invariant and instance-specific features, enabling class-consistent instance generation across varying dimensions without problem-specific instance generators. Second, we introduce LLM-based automatic search operator generation into PAP construction, extending the search space from parameter tuning of predefined templates to operator-level algorithm design. We evaluate DACMO on four representative MOBOP classes to demonstrate its effectiveness as a general-purpose PAP construction method: the multi-objective match max problem~(MMMP), the multi-objective knapsack problem~(MKP), the multi-objective contamination control problem (MCCP), and the multi-objective complementary influence maximization problem~(MCIMP). Experimental results show that DACMO can be directly applied to all four problem classes without modification, outperforms PAPs built from classic MOEA templates, and achieves performance comparable to a privileged state-of-the-art baseline that relies on manually designed problem-specific instance generators, while outperforming it on two of the four evaluated problem classes.

Xunzhi Xiang, Zixuan Duan, Guiyu Zhang et al.

Distilled autoregressive diffusion models facilitate real-time short video synthesis but suffer from severe error accumulation during long-sequence generation. While existing Test-Time Optimization (TTO) methods prove effective for images or short clips, we identify that they fail to mitigate drift in extended sequences due to unstable reward landscapes and the hypersensitivity of distilled parameters. To overcome these limitations, we introduce Test-Time Correction (TTC), a training-free alternative. Specifically, TTC utilizes the initial frame as a stable reference anchor to calibrate intermediate stochastic states along the sampling trajectory. Extensive experiments demonstrate that our method seamlessly integrates with various distilled models, extending generation lengths with negligible overhead while matching the quality of resource-intensive training-based methods on 30-second benchmarks.

Yan Li, Weiwei Guo, Xue Yang et al.

In recent years, aerial object detection has been increasingly pivotal in various earth observation applications. However, current algorithms are limited to detecting a set of pre-defined object categories, demanding sufficient annotated training samples, and fail to detect novel object categories. In this paper, we put forth a novel formulation of the aerial object detection problem, namely open-vocabulary aerial object detection (OVAD), which can detect objects beyond training categories without costly collecting new labeled data. We propose CastDet, a CLIP-activated student-teacher detection framework that serves as the first OVAD detector specifically designed for the challenging aerial scenario, where objects often exhibit weak appearance features and arbitrary orientations. Our framework integrates a robust localization teacher along with several box selection strategies to generate high-quality proposals for novel objects. Additionally, the RemoteCLIP model is adopted as an omniscient teacher, which provides rich knowledge to enhance classification capabilities for novel categories. A dynamic label queue is devised to maintain high-quality pseudo-labels during training. By doing so, the proposed CastDet boosts not only novel object proposals but also classification. Furthermore, we extend our approach from horizontal OVAD to oriented OVAD with tailored algorithm designs to effectively manage bounding box representation and pseudo-label generation. Extensive experiments for both tasks on multiple existing aerial object detection datasets demonstrate the effectiveness of our approach. The code is available at https://github.com/lizzy8587/CastDet.

Mingxin Liu, Shuran Ma, Shibei Meng et al.

While generative video models have achieved remarkable visual fidelity, their capacity to internalize and reason over implicit world rules remains a critical yet under-explored frontier. To bridge this gap, we present RISE-Video, a pioneering reasoning-oriented benchmark for Text-Image-to-Video (TI2V) synthesis that shifts the evaluative focus from surface-level aesthetics to deep cognitive reasoning. RISE-Video comprises 467 meticulously human-annotated samples spanning eight rigorous categories, providing a structured testbed for probing model intelligence across diverse dimensions, ranging from commonsense and spatial dynamics to specialized subject domains. Our framework introduces a multi-dimensional evaluation protocol consisting of four metrics: \textit{Reasoning Alignment}, \textit{Temporal Consistency}, \textit{Physical Rationality}, and \textit{Visual Quality}. To further support scalable evaluation, we propose an automated pipeline leveraging Large Multimodal Models (LMMs) to emulate human-centric assessment. Extensive experiments on 11 state-of-the-art TI2V models reveal pervasive deficiencies in simulating complex scenarios under implicit constraints, offering critical insights for the advancement of future world-simulating generative models.

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.

Shaofeng Zhang, Jinfa Huang, Qiang Zhou et al.

Image outpainting aims to generate the content of an input sub-image beyond its original boundaries. It is an important task in content generation yet remains an open problem for generative models. This paper pushes the technical frontier of image outpainting in two directions that have not been resolved in literature: 1) outpainting with arbitrary and continuous multiples (without restriction), and 2) outpainting in a single step (even for large expansion multiples). Moreover, we develop a method that does not depend on a pre-trained backbone network, which is in contrast commonly required by the previous SOTA outpainting methods. The arbitrary multiple outpainting is achieved by utilizing randomly cropped views from the same image during training to capture arbitrary relative positional information. Specifically, by feeding one view and positional embeddings as queries, we can reconstruct another view. At inference, we generate images with arbitrary expansion multiples by inputting an anchor image and its corresponding positional embeddings. The one-step outpainting ability here is particularly noteworthy in contrast to previous methods that need to be performed for $N$ times to obtain a final multiple which is $N$ times of its basic and fixed multiple. We evaluate the proposed approach (called PQDiff as we adopt a diffusion-based generator as our embodiment, under our proposed \textbf{P}ositional \textbf{Q}uery scheme) on public benchmarks, demonstrating its superior performance over state-of-the-art approaches. Specifically, PQDiff achieves state-of-the-art FID scores on the Scenery (\textbf{21.512}), Building Facades (\textbf{25.310}), and WikiArts (\textbf{36.212}) datasets. Furthermore, under the 2.25x, 5x and 11.7x outpainting settings, PQDiff only takes \textbf{40.6\%}, \textbf{20.3\%} and \textbf{10.2\%} of the time of the benchmark state-of-the-art (SOTA) method.

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/.

Qiang Zhou, Shaofeng Zhang, Nianzu Yang et al.

Existing text-to-video (T2V) models often struggle with generating videos with sufficiently pronounced or complex actions. A key limitation lies in the text prompt's inability to precisely convey intricate motion details. To address this, we propose a novel framework, MVideo, designed to produce long-duration videos with precise, fluid actions. MVideo overcomes the limitations of text prompts by incorporating mask sequences as an additional motion condition input, providing a clearer, more accurate representation of intended actions. Leveraging foundational vision models such as GroundingDINO and SAM2, MVideo automatically generates mask sequences, enhancing both efficiency and robustness. Our results demonstrate that, after training, MVideo effectively aligns text prompts with motion conditions to produce videos that simultaneously meet both criteria. This dual control mechanism allows for more dynamic video generation by enabling alterations to either the text prompt or motion condition independently, or both in tandem. Furthermore, MVideo supports motion condition editing and composition, facilitating the generation of videos with more complex actions. MVideo thus advances T2V motion generation, setting a strong benchmark for improved action depiction in current video diffusion models. Our project page is available at https://mvideo-v1.github.io/.

Xunzhi Xiang, Xingye Tian, Guiyu Zhang et al.

Variational autoencoders (VAEs) typically encode images into a compact latent space, reducing computational cost but introducing an optimization dilemma: a higher-dimensional latent space improves reconstruction fidelity but often hampers generative performance. Recent methods attempt to address this dilemma by regularizing high-dimensional latent spaces using external vision foundation models (VFMs). However, it remains unclear how high-dimensional VAE latents affect the optimization of generative models. To our knowledge, our analysis is the first to reveal that redundant high-frequency components in high-dimensional latent spaces hinder the training convergence of diffusion models and, consequently, degrade generation quality. To alleviate this problem, we propose a spectral self-regularization strategy to suppress redundant high-frequency noise while simultaneously preserving reconstruction quality. The resulting Denoising-VAE, a ViT-based autoencoder that does not rely on VFMs, produces cleaner, lower-noise latents, leading to improved generative quality and faster optimization convergence. We further introduce a spectral alignment strategy to facilitate the optimization of Denoising-VAE-based generative models. Our complete method enables diffusion models to converge approximately 2$\times$ faster than with SD-VAE, while achieving state-of-the-art reconstruction quality (rFID = 0.28, PSNR = 27.26) and competitive generation performance (gFID = 1.82) on the ImageNet 256$\times$256 benchmark.

Shenghai Yuan, Jinfa Huang, Yongqi Xu et al.

We propose a novel text-to-video (T2V) generation benchmark, ChronoMagic-Bench, to evaluate the temporal and metamorphic capabilities of the T2V models (e.g. Sora and Lumiere) in time-lapse video generation. In contrast to existing benchmarks that focus on visual quality and textual relevance of generated videos, ChronoMagic-Bench focuses on the model's ability to generate time-lapse videos with significant metamorphic amplitude and temporal coherence. The benchmark probes T2V models for their physics, biology, and chemistry capabilities, in a free-form text query. For these purposes, ChronoMagic-Bench introduces 1,649 prompts and real-world videos as references, categorized into four major types of time-lapse videos: biological, human-created, meteorological, and physical phenomena, which are further divided into 75 subcategories. This categorization comprehensively evaluates the model's capacity to handle diverse and complex transformations. To accurately align human preference with the benchmark, we introduce two new automatic metrics, MTScore and CHScore, to evaluate the videos' metamorphic attributes and temporal coherence. MTScore measures the metamorphic amplitude, reflecting the degree of change over time, while CHScore assesses the temporal coherence, ensuring the generated videos maintain logical progression and continuity. Based on ChronoMagic-Bench, we conduct comprehensive manual evaluations of ten representative T2V models, revealing their strengths and weaknesses across different categories of prompts, and providing a thorough evaluation framework that addresses current gaps in video generation research. Moreover, we create a large-scale ChronoMagic-Pro dataset, containing 460k high-quality pairs of 720p time-lapse videos and detailed captions ensuring high physical pertinence and large metamorphic amplitude. [Homepage](https://pku-yuangroup.github.io/ChronoMagic-Bench/).