Bo Peng

Bo Peng, Kaiwen Wu, Sirui Chen et al. · tencent-ai

Causal discovery from observational data remains challenging due to the fundamental limitations of purely statistical methods, such as statistical distinguishability within equivalence classes and sensitivity to finite sample sizes. While large language models (LLMs) offer a promising source of domain knowledge to complement statistical inference, existing LLM-augmented methods are vulnerable to LLM errors and incur high token costs. Moreover, reliance on a single data-centric algorithm can make results sensitive to algorithm-specific biases. To address these limitations, we propose CauTion, a framework that reliably integrates LLM domain knowledge into an ensemble of statistical causal discovery algorithms through consensus filtering and LLM reliability estimation. CauTion proceeds in three stages. First, an algorithm ensemble utilizes a consensus voting to resolve up to 96% of edges on which algorithms agree, achieving near-perfect accuracy on the filtered consensus edges. Second, a trust-calibrated arbitration mechanism estimates the relative reliability of the LLM and the algorithms via an annotation-free trust calibration procedure, which is then utilized to govern a trust-weighted voting process that restricts LLM arbitration exclusively to edges with unreliable algorithmic evidence. Third, a cycle repair step is applied to guarantee the final causal graph is validly acyclic. Experiments on six datasets demonstrate that CauTion consistently outperforms both data-centric and LLM-augmented baselines, with larger gains on larger graphs and strong robustness to LLM errors. Code is available at https://github.com/OpenCausaLab/CauTion.

Mufeng Yao, Jinlong Peng, Qingdong He et al.

Multiple object tracking (MOT) from unmanned aerial vehicle (UAV) platforms requires efficient motion modeling. This is because UAV-MOT faces both local object motion and global camera motion. Motion blur also increases the difficulty of detecting large moving objects. Previous UAV motion modeling approaches either focus only on local motion or ignore motion blurring effects, thus limiting their tracking performance and speed. To address these issues, we propose the Motion Mamba Module, which explores both local and global motion features through cross-correlation and bi-directional Mamba Modules for better motion modeling. To address the detection difficulties caused by motion blur, we also design motion margin loss to effectively improve the detection accuracy of motion blurred objects. Based on the Motion Mamba module and motion margin loss, our proposed MM-Tracker surpasses the state-of-the-art in two widely open-source UAV-MOT datasets. Code will be available.

Bowen Peng, Bo Peng, Jie Zhou et al. · tencent-ai

Deep Neural Networks (DNNs) based Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) systems have shown to be highly vulnerable to adversarial perturbations that are deliberately designed yet almost imperceptible but can bias DNN inference when added to targeted objects. This leads to serious safety concerns when applying DNNs to high-stake SAR ATR applications. Therefore, enhancing the adversarial robustness of DNNs is essential for implementing DNNs to modern real-world SAR ATR systems. Toward building more robust DNN-based SAR ATR models, this article explores the domain knowledge of SAR imaging process and proposes a novel Scattering Model Guided Adversarial Attack (SMGAA) algorithm which can generate adversarial perturbations in the form of electromagnetic scattering response (called adversarial scatterers). The proposed SMGAA consists of two parts: 1) a parametric scattering model and corresponding imaging method and 2) a customized gradient-based optimization algorithm. First, we introduce the effective Attributed Scattering Center Model (ASCM) and a general imaging method to describe the scattering behavior of typical geometric structures in the SAR imaging process. By further devising several strategies to take the domain knowledge of SAR target images into account and relax the greedy search procedure, the proposed method does not need to be prudentially finetuned, but can efficiently to find the effective ASCM parameters to fool the SAR classifiers and facilitate the robust model training. Comprehensive evaluations on the MSTAR dataset show that the adversarial scatterers generated by SMGAA are more robust to perturbations and transformations in the SAR processing chain than the currently studied attacks, and are effective to construct a defensive model against the malicious scatterers.

Bo Peng, Wei Xiang, Yue Jiang et al.

This paper presents the summary report on our DFGC 2022 competition. The DeepFake is rapidly evolving, and realistic face-swaps are becoming more deceptive and difficult to detect. On the contrary, methods for detecting DeepFakes are also improving. There is a two-party game between DeepFake creators and defenders. This competition provides a common platform for benchmarking the game between the current state-of-the-arts in DeepFake creation and detection methods. The main research question to be answered by this competition is the current state of the two adversaries when competed with each other. This is the second edition after the last year's DFGC 2021, with a new, more diverse video dataset, a more realistic game setting, and more reasonable evaluation metrics. With this competition, we aim to stimulate research ideas for building better defenses against the DeepFake threats. We also release our DFGC 2022 dataset contributed by both our participants and ourselves to enrich the DeepFake data resources for the research community (https://github.com/NiCE-X/DFGC-2022).

Xianyun Sun, Beibei Dong, Caiyong Wang et al.

Deep-learning based face-swap videos, also known as deep fakes, are becoming more and more realistic and deceiving. The malicious usage of these face-swap videos has caused wide concerns. The research community has been focusing on the automatic detection of these fake videos, but the assessment of their visual realism, as perceived by human eyes, is still an unexplored dimension. Visual realism assessment, or VRA, is essential for assessing the potential impact that may be brought by a specific face-swap video, and it is also important as a quality assessment metric to compare different face-swap methods. In this paper, we make a small step towards this new VRA direction by building a benchmark for evaluating the effectiveness of different automatic VRA models, which range from using traditional hand-crafted features to different kinds of deep-learning features. The evaluations are based on a recent competition dataset named DFGC 2022, which contains 1400 diverse face-swap videos that are annotated with Mean Opinion Scores (MOS) on visual realism. Comprehensive experiment results using 11 models and 3 protocols are shown and discussed. We demonstrate the feasibility of devising effective VRA models for assessing face-swap videos and methods. The particular usefulness of existing deepfake detection features for VRA is also noted. The code can be found at https://github.com/XianyunSun/VRA.git.

Qi Li, Weining Wang, Shuangjun Du et al.

Face swapping has witnessed significant progress in recent years, largely driven by advances in deep generative models such as GANs and diffusion models.Despite these advances, existing methods remain fragmented across different paradigms, and their evaluation is highly inconsistent due to the lack of standardized datasets and protocols. Moreover, prior surveys primarily focus on broader deepfake generation or detection, leaving face swapping insufficiently studied as a standalone problem. In this paper, we present a comprehensive survey and benchmark for face swapping. We provide a structured review of existing methods, organizing them into five major paradigms and systematically analyzing their design principles, strengths, and limitations. To enable fair and controlled evaluation, we introduce CASIA FaceSwapping, a high-quality benchmark with balanced demographic distributions and explicit attribute variations, and establish standardized protocols to assess the robustness of different face swapping methods. Extensive experiments on representative approaches yield new insights into the performance characteristics and limitations of current techniques. Overall, our work provides a unified perspective and a principled evaluation framework to facilitate the development of more robust and controllable face swapping methods. More results can be found at https://github.com/CASIA-NLPRAI/face-swapping-survey.

Bo Peng, Jie Lu, Guangquan Zhang et al.

Aiming at identifying unexpected inputs from unknown classes, out-of-distribution (OOD) detection has emerged as a pivotal approach to enhancing the reliability of machine learning models. This paper focuses on the burgeoning paradigm of post-hoc OOD detection with pre-trained vision-language models (VLMs), where a popular pipeline is to detect OOD inputs by examining their affinities between ID labels and negative labels, i.e., those semantically different from ID labels. Due to the unavailability of target OOD labels, existing works predominantly rely on heuristic rules to mine negative labels from unlabeled wild corpus data. Despite the empirical success, we argue that the power of VLM-based OOD detection has yet to be fully unleashed since the notorious false negative problem is far from addressed in the literature. With this motivation, we are interested in addressing the challenge of mining true negative labels for OOD scoring. To this end, we develop a theoretical framework for correcting the sampling bias of negatives labels by indirectly approximating the distribution of negative labels. Perhaps surprisingly, we show that the debiased negative mining can be naturally converted into Monte-Carlo sampling based on ID labels and the unlabeled wild corpus data. Extensive experiments empirically manifest that our method establishes a new state-of-the-art in a variety of OOD detection setups. Code is publicly available at \href{https://github.com/60pen9/Debiased-Negative-Mining-Improves-OOD-Detection-with-Pre-trained-VLMs}{\textcolor{red}{here}}.

Bo Peng, Jun Hu, Jingtao Zhou et al.

Recently, many works have been proposed to utilize the neural radiance field for novel view synthesis of human performers. However, most of these methods require hours of training, making them difficult for practical use. To address this challenging problem, we propose IntrinsicNGP, which can train from scratch and achieve high-fidelity results in few minutes with videos of a human performer. To achieve this target, we introduce a continuous and optimizable intrinsic coordinate rather than the original explicit Euclidean coordinate in the hash encoding module of instant-NGP. With this novel intrinsic coordinate, IntrinsicNGP can aggregate inter-frame information for dynamic objects with the help of proxy geometry shapes. Moreover, the results trained with the given rough geometry shapes can be further refined with an optimizable offset field based on the intrinsic coordinate.Extensive experimental results on several datasets demonstrate the effectiveness and efficiency of IntrinsicNGP. We also illustrate our approach's ability to edit the shape of reconstructed subjects.

Yueming Lyu, Kang Zhao, Bo Peng et al.

Text-guided image editing faces significant challenges when considering training and inference flexibility. Much literature collects large amounts of annotated image-text pairs to train text-conditioned generative models from scratch, which is expensive and not efficient. After that, some approaches that leverage pre-trained vision-language models have been proposed to avoid data collection, but they are limited by either per text-prompt optimization or inference-time hyper-parameters tuning. To address these issues, we investigate and identify a specific space, referred to as CLIP DeltaSpace, where the CLIP visual feature difference of two images is semantically aligned with the CLIP textual feature difference of their corresponding text descriptions. Based on DeltaSpace, we propose a novel framework called DeltaEdit, which maps the CLIP visual feature differences to the latent space directions of a generative model during the training phase, and predicts the latent space directions from the CLIP textual feature differences during the inference phase. And this design endows DeltaEdit with two advantages: (1) text-free training; (2) generalization to various text prompts for zero-shot inference. Extensive experiments validate the effectiveness and versatility of DeltaEdit with different generative models, including both the GAN model and the diffusion model, in achieving flexible text-guided image editing. Code is available at https://github.com/Yueming6568/DeltaEdit.

Xiaofan Li, Bo Peng, Jie Hu et al.

Unsupervised skin lesion segmentation offers several benefits, including conserving expert human resources, reducing discrepancies due to subjective human labeling, and adapting to novel environments. However, segmenting dermoscopic images without manual labeling guidance presents significant challenges due to dermoscopic image artifacts such as hair noise, blister noise, and subtle edge differences. To address these challenges, we introduce an innovative Uncertainty Self-Learning Network (USL-Net) designed for skin lesion segmentation. The USL-Net can effectively segment a range of lesions, eliminating the need for manual labeling guidance. Initially, features are extracted using contrastive learning, followed by the generation of Class Activation Maps (CAMs) as saliency maps using these features. The different CAM locations correspond to the importance of the lesion region based on their saliency. High-saliency regions in the map serve as pseudo-labels for lesion regions while low-saliency regions represent the background. However, intermediate regions can be hard to classify, often due to their proximity to lesion edges or interference from hair or blisters. Rather than risk potential pseudo-labeling errors or learning confusion by forcefully classifying these regions, we consider them as uncertainty regions, exempting them from pseudo-labeling and allowing the network to self-learn. Further, we employ connectivity detection and centrality detection to refine foreground pseudo-labels and reduce noise-induced errors. The application of cycle refining enhances performance further. Our method underwent thorough experimental validation on the ISIC-2017, ISIC-2018, and PH2 datasets, demonstrating that its performance is on par with weakly supervised and supervised methods, and exceeds that of other existing unsupervised methods.

Weinan Guan, Wei Wang, Jing Dong et al.

Detecting maliciously falsified facial images and videos has attracted extensive attention from digital-forensics and computer-vision communities. An important topic in manipulation detection is the localization of the fake regions. Previous work related to forgery detection mostly focuses on the entire faces. However, recent forgery methods have developed to edit important facial components while maintaining others unchanged. This drives us to not only focus on the forgery detection but also fine-grained falsified region segmentation. In this paper, we propose a collaborative feature learning approach to simultaneously detect manipulation and segment the falsified components. With the collaborative manner, detection and segmentation can boost each other efficiently. To enable our study of forgery detection and segmentation, we build a facial forgery dataset consisting of both entire and partial face forgeries with their pixel-level manipulation ground-truth. Experiment results have justified the mutual promotion between forgery detection and manipulated region segmentation. The overall performance of the proposed approach is better than the state-of-the-art detection or segmentation approaches. The visualization results have shown that our proposed model always captures the artifacts on facial regions, which is more reasonable.

Gyeongsik Moon, Shunsuke Saito, Weipeng Xu et al.

The two-hand interaction is one of the most challenging signals to analyze due to the self-similarity, complicated articulations, and occlusions of hands. Although several datasets have been proposed for the two-hand interaction analysis, all of them do not achieve 1) diverse and realistic image appearances and 2) diverse and large-scale groundtruth (GT) 3D poses at the same time. In this work, we propose Re:InterHand, a dataset of relighted 3D interacting hands that achieve the two goals. To this end, we employ a state-of-the-art hand relighting network with our accurately tracked two-hand 3D poses. We compare our Re:InterHand with existing 3D interacting hands datasets and show the benefit of it. Our Re:InterHand is available in https://mks0601.github.io/ReInterHand/.

Bo Peng, Jun Hu, Jingtao Zhou et al.

In this paper, we propose SelfNeRF, an efficient neural radiance field based novel view synthesis method for human performance. Given monocular self-rotating videos of human performers, SelfNeRF can train from scratch and achieve high-fidelity results in about twenty minutes. Some recent works have utilized the neural radiance field for dynamic human reconstruction. However, most of these methods need multi-view inputs and require hours of training, making it still difficult for practical use. To address this challenging problem, we introduce a surface-relative representation based on multi-resolution hash encoding that can greatly improve the training speed and aggregate inter-frame information. Extensive experimental results on several different datasets demonstrate the effectiveness and efficiency of SelfNeRF to challenging monocular videos.

Tingyu Zhao, Bo Peng, Yuan Sun et al.

Recently, advancements in deep learning-based superpixel segmentation methods have brought about improvements in both the efficiency and the performance of segmentation. However, a significant challenge remains in generating superpixels that strictly adhere to object boundaries while conveying rich visual significance, especially when cross-surface color correlations may interfere with objects. Drawing inspiration from neural structure and visual mechanisms, we propose a biological network architecture comprising an Enhanced Screening Module (ESM) and a novel Boundary-Aware Label (BAL) for superpixel segmentation. The ESM enhances semantic information by simulating the interactive projection mechanisms of the visual cortex. Additionally, the BAL emulates the spatial frequency characteristics of visual cortical cells to facilitate the generation of superpixels with strong boundary adherence. We demonstrate the effectiveness of our approach through evaluations on both the BSDS500 dataset and the NYUv2 dataset.

Bo Peng, YuanJie Lyu, PengGang Qin et al.

Accurately predicting future events is fundamental to content understanding and decision-making across various domains. While prior research has primarily focused on text or short-video scenarios, long-video event prediction, characterized by vast multimodal context and more complex narratives, remains underexplored. Meanwhile, although recent Long-Video Language Models (LVLMs), built on Large Language Models (LLMs) and Vision-Language Models (VLMs), have shown promise in long-video question answering and summarization, they struggle to generalize to event prediction, as they can neither precisely extract event-related details nor perform fine-grained analysis of event development. To address this gap, we propose VISTA, a multi-level event semantics mining framework for long-video event prediction. Initially, VISTA applies a character-centric visual prompt to precisely extract event-related visual details, enhancing detail-level semantics; subsequently, it employs a knowledge-enhanced iterative retrieval strategy, guiding the LLM to progressively construct logically coherent event chains, thereby improving event-level narratives; ultimately, VISTA adopts a human-like propose-then-retrieve strategy to generate diverse future-oriented proposals and integrate multi-level clues, producing robust and accurate predictions. Extensive experiments on real-world datasets validate the effectiveness of VISTA for long-video event prediction.

Ziqi Chen, Bo Peng, Srinivasan Parthasarathy et al.

Ligand-based drug design aims to identify novel drug candidates of similar shapes with known active molecules. In this paper, we formulated an in silico shape-conditioned molecule generation problem to generate 3D molecule structures conditioned on the shape of a given molecule. To address this problem, we developed a translation- and rotation-equivariant shape-guided generative model ShapeMol. ShapeMol consists of an equivariant shape encoder that maps molecular surface shapes into latent embeddings, and an equivariant diffusion model that generates 3D molecules based on these embeddings. Experimental results show that ShapeMol can generate novel, diverse, drug-like molecules that retain 3D molecular shapes similar to the given shape condition. These results demonstrate the potential of ShapeMol in designing drug candidates of desired 3D shapes binding to protein target pockets.

Yize Zhang, Meiqi Chen, Sirui Chen et al.

Causal thinking enables humans to understand not just what is seen, but why it happens. To replicate this capability in modern AI systems, we introduce the task of visual causal discovery. It requires models to infer cause-and-effect relations among visual entities across diverse scenarios instead of merely perceiving their presence. To this end, we first construct the Visual Causal Graph dataset (VCG-32K), a large-scale collection of over 32,000 images annotated with entity-level causal graphs, and further develop CauSight, a novel vision-language model to perform visual causal discovery through causally aware reasoning. Our training recipe integrates three components: (1) training data curation from VCG-32K, (2) Tree-of-Causal-Thought (ToCT) for synthesizing reasoning trajectories, and (3) reinforcement learning with a designed causal reward to refine the reasoning policy. Experiments show that CauSight outperforms GPT-4.1 on visual causal discovery, achieving over a threefold performance boost (21% absolute gain). Our code, model, and dataset are fully open-sourced at project page: https://github.com/OpenCausaLab/CauSight.

Songlin Yang, Wei Wang, Bo Peng et al.

GAN inversion has been exploited in many face manipulation tasks, but 2D GANs often fail to generate multi-view 3D consistent images. The encoders designed for 2D GANs are not able to provide sufficient 3D information for the inversion and editing. Therefore, 3D-aware GAN inversion is proposed to increase the 3D editing capability of GANs. However, the 3D-aware GAN inversion remains under-explored. To tackle this problem, we propose a 3D-aware (3Da) encoder for GAN inversion and face editing based on the powerful StyleNeRF model. Our proposed 3Da encoder combines a parametric 3D face model with a learnable detail representation model to generate geometry, texture and view direction codes. For more flexible face manipulation, we then design a dual-branch StyleFlow module to transfer the StyleNeRF codes with disentangled geometry and texture flows. Extensive experiments demonstrate that we realize 3D consistent face manipulation in both facial attribute editing and texture transfer. Furthermore, for video editing, we make the sequence of frame codes share a common canonical manifold, which improves the temporal consistency of the edited attributes.

Yuanwei Hu, Bo Peng, Yadan Luo et al.

Out-of-distribution (OOD) detection has emerged as a popular technique to enhance the reliability of machine learning models by identifying unexpected inputs from unknown classes. Recent progress in pre-trained vision-language models (VLMs) has enabled zero-shot OOD detection without access to in-distribution (ID) training data; in this setting, existing methods commonly treat text embeddings of class names as class prototypes. In this paper, we challenge the widely adopted text-as-prototype paradigm by theoretically showing that off-the-shelf textual prototypes are generally misaligned with the optimal visual prototypes, yielding an intrinsic modality gap that cannot be eliminated by prompt engineering alone. To mitigate this gap under the post-hoc constraint, this paper presents an online pseudo-supervised framework that directly learns class prototypes in the visual feature space using unlabeled test-time data streams and soft predictions from the pre-trained VLMs. We provide theoretical guarantees for the convergence of the online optimization procedure. Extensive experiments empirically demonstrate that our method achieves a new state of the art across a variety of OOD detection setups.

Yueming Lyu, Yue Jiang, Ziwen He et al.

The privacy and security of face data on social media are facing unprecedented challenges as it is vulnerable to unauthorized access and identification. A common practice for solving this problem is to modify the original data so that it could be protected from being recognized by malicious face recognition (FR) systems. However, such ``adversarial examples'' obtained by existing methods usually suffer from low transferability and poor image quality, which severely limits the application of these methods in real-world scenarios. In this paper, we propose a 3D-Aware Adversarial Makeup Generation GAN (3DAM-GAN). which aims to improve the quality and transferability of synthetic makeup for identity information concealing. Specifically, a UV-based generator consisting of a novel Makeup Adjustment Module (MAM) and Makeup Transfer Module (MTM) is designed to render realistic and robust makeup with the aid of symmetric characteristics of human faces. Moreover, a makeup attack mechanism with an ensemble training strategy is proposed to boost the transferability of black-box models. Extensive experiment results on several benchmark datasets demonstrate that 3DAM-GAN could effectively protect faces against various FR models, including both publicly available state-of-the-art models and commercial face verification APIs, such as Face++, Baidu and Aliyun.

Bo Peng, Sirui Chen, Lei Xu et al.

Causal discovery is fundamental to scientific understanding and reliable decision-making. Existing approaches face critical limitations: purely data-driven methods suffer from statistical indistinguishability and modeling assumptions, while recent LLM-based methods either ignore statistical evidence or incorporate unverified priors that can mislead result. To this end, we propose CauScientist, a collaborative framework that synergizes LLMs as hypothesis-generating "data scientists" with probabilistic statistics as rigorous "verifiers". CauScientist employs hybrid initialization to select superior starting graphs, iteratively refines structures through LLM-proposed modifications validated by statistical criteria, and maintains error memory to guide efficient search space. Experiments demonstrate that CauScientist substantially outperforms purely data-driven baselines, achieving up to 53.8% F1 score improvement and enhancing recall from 35.0% to 100.0%. Notably, while standalone LLM performance degrades with graph complexity, CauScientist reduces structural hamming distance (SHD) by 44.0% compared to Qwen3-32B on 37-node graphs. Our project page is at https://github.com/OpenCausaLab/CauScientist.

Bo Peng, Sirui Chen, Jiaguo Tian et al.

Causal discovery is essential for advancing data-driven fields such as scientific AI and data analysis, yet existing approaches face significant time- and space-efficiency bottlenecks when scaling to large graphs. To address this challenge, we present CauScale, a neural architecture designed for efficient causal discovery that scales inference to graphs with up to 1000 nodes. CauScale improves time efficiency via a reduction unit that compresses data embeddings and improves space efficiency by adopting tied attention weights to avoid maintaining axis-specific attention maps. To keep high causal discovery accuracy, CauScale adopts a two-stream design: a data stream extracts relational evidence from high-dimensional observations, while a graph stream integrates statistical graph priors and preserves key structural signals. CauScale successfully scales to 500-node graphs during training, where prior work fails due to space limitations. Across testing data with varying graph scales and causal mechanisms, CauScale achieves 99.6% mAP on in-distribution data and 84.4% on out-of-distribution data, while delivering 4-13,000 times inference speedups over prior methods. Our project page is at https://github.com/OpenCausaLab/CauScale.

Yueming Lyu, Yue Jiang, Bo Peng et al.

Artistic style transfer aims to transfer the style of an artwork to a photograph while maintaining its original overall content. Many prior works focus on designing various transfer modules to transfer the style statistics to the content image. Although effective, ignoring the clear disentanglement of the content features and the style features from the first beginning, they have difficulty in balancing between content preservation and style transferring. To tackle this problem, we propose a novel information disentanglement method, named InfoStyler, to capture the minimal sufficient information for both content and style representations from the pre-trained encoding network. InfoStyler formulates the disentanglement representation learning as an information compression problem by eliminating style statistics from the content image and removing the content structure from the style image. Besides, to further facilitate disentanglement learning, a cross-domain Information Bottleneck (IB) learning strategy is proposed by reconstructing the content and style domains. Extensive experiments demonstrate that our InfoStyler can synthesize high-quality stylized images while balancing content structure preservation and style pattern richness.

Bowen Peng, Jianyue Xie, Bo Peng et al.

The deep neural networks (DNNs) have freed the synthetic aperture radar automatic target recognition (SAR ATR) from expertise-based feature designing and demonstrated superiority over conventional solutions. There has been shown the unique deficiency of ground vehicle benchmarks in shapes of strong background correlation results in DNNs overfitting the clutter and being non-robust to unfamiliar surroundings. However, the gap between fixed background model training and varying background application remains underexplored. Inspired by contrastive learning, this letter proposes a solution called Contrastive Feature Alignment (CFA) aiming to learn invariant representation for robust recognition. The proposed method contributes a mixed clutter variants generation strategy and a new inference branch equipped with channel-weighted mean square error (CWMSE) loss for invariant representation learning. In specific, the generation strategy is delicately designed to better attract clutter-sensitive deviation in feature space. The CWMSE loss is further devised to better contrast this deviation and align the deep features activated by the original images and corresponding clutter variants. The proposed CFA combines both classification and CWMSE losses to train the model jointly, which allows for the progressive learning of invariant target representation. Extensive evaluations on the MSTAR dataset and six DNN models prove the effectiveness of our proposal. The results demonstrated that the CFA-trained models are capable of recognizing targets among unfamiliar surroundings that are not included in the dataset, and are robust to varying signal-to-clutter ratios.

Bo Peng, Hongchen Liu, Hang Zhou et al.

Earth observation satellites have been continuously monitoring the earth environment for years at different locations and spectral bands with different modalities. Due to complex satellite sensing conditions (e.g., weather, cloud, atmosphere, orbit), some observations for certain modalities, bands, locations, and times may not be available. The MultiEarth Matrix Completion Challenge in CVPR 2022 [1] provides the multimodal satellite data for addressing such data sparsity challenges with the Amazon Rainforest as the region of interest. This work proposes an adaptive real-time multimodal regression and generation framework and achieves superior performance on unseen test queries in this challenge with an LPIPS of 0.2226, a PSNR of 123.0372, and an SSIM of 0.6347.

Yuchuan Gou, Bo Peng, Hongchen Liu et al.

The MultiEarth 2022 Image-to-Image Translation challenge provides a well-constrained test bed for generating the corresponding RGB Sentinel-2 imagery with the given Sentinel-1 VV & VH imagery. In this challenge, we designed various generation models and found the SPADE [1] and pix2pixHD [2] models could perform our best results. In our self-evaluation, the SPADE-2 model with L1-loss can achieve 0.02194 MAE score and 31.092 PSNR dB. In our final submission, the best model can achieve 0.02795 MAE score ranked No.1 on the leader board.

Zhenchen Tang, Zichuan Wang, Bo Peng et al.

With the rapid development of generative technologies, AI-Generated Images (AIGIs) have been widely applied in various aspects of daily life. However, due to the immaturity of the technology, the quality of the generated images varies, so it is important to develop quality assessment techniques for the generated images. Although some models have been proposed to assess the quality of generated images, they are inadequate when faced with the ever-increasing and diverse categories of generated images. Consequently, the development of more advanced and effective models for evaluating the quality of generated images is urgently needed. Recent research has explored the significant potential of the visual language model CLIP in image quality assessment, finding that it performs well in evaluating the quality of natural images. However, its application to generated images has not been thoroughly investigated. In this paper, we build on this idea and further explore the potential of CLIP in evaluating the quality of generated images. We design CLIP-AGIQA, a CLIP-based regression model for quality assessment of generated images, leveraging rich visual and textual knowledge encapsulated in CLIP. Particularly, we implement multi-category learnable prompts to fully utilize the textual knowledge in CLIP for quality assessment. Extensive experiments on several generated image quality assessment benchmarks, including AGIQA-3K and AIGCIQA2023, demonstrate that CLIP-AGIQA outperforms existing IQA models, achieving excellent results in evaluating the quality of generated images.

Dongze Li, Kang Zhao, Wei Wang et al.

Talking head synthesis is a practical technique with wide applications. Current Neural Radiance Field (NeRF) based approaches have shown their superiority on driving one-shot talking heads with videos or signals regressed from audio. However, most of them failed to take the audio as driven information directly, unable to enjoy the flexibility and availability of speech. Since mapping audio signals to face deformation is non-trivial, we design a Single-Shot Speech-Driven Neural Radiance Field (S^3D-NeRF) method in this paper to tackle the following three difficulties: learning a representative appearance feature for each identity, modeling motion of different face regions with audio, and keeping the temporal consistency of the lip area. To this end, we introduce a Hierarchical Facial Appearance Encoder to learn multi-scale representations for catching the appearance of different speakers, and elaborate a Cross-modal Facial Deformation Field to perform speech animation according to the relationship between the audio signal and different face regions. Moreover, to enhance the temporal consistency of the important lip area, we introduce a lip-sync discriminator to penalize the out-of-sync audio-visual sequences. Extensive experiments have shown that our S^3D-NeRF surpasses previous arts on both video fidelity and audio-lip synchronization.

Songlin Yang, Wei Wang, Chenye Xu et al.

Face anti-spoofing aims to discriminate the spoofing face images (e.g., printed photos) from live ones. However, adversarial examples greatly challenge its credibility, where adding some perturbation noise can easily change the predictions. Previous works conducted adversarial attack methods to evaluate the face anti-spoofing performance without any fine-grained analysis that which model architecture or auxiliary feature is vulnerable to the adversary. To handle this problem, we propose a novel framework to expose the fine-grained adversarial vulnerability of the face anti-spoofing models, which consists of a multitask module and a semantic feature augmentation (SFA) module. The multitask module can obtain different semantic features for further evaluation, but only attacking these semantic features fails to reflect the discrimination-related vulnerability. We then design the SFA module to introduce the data distribution prior for more discrimination-related gradient directions for generating adversarial examples. Comprehensive experiments show that SFA module increases the attack success rate by nearly 40$\%$ on average. We conduct this fine-grained adversarial analysis on different annotations, geometric maps, and backbone networks (e.g., Resnet network). These fine-grained adversarial examples can be used for selecting robust backbone networks and auxiliary features. They also can be used for adversarial training, which makes it practical to further improve the accuracy and robustness of the face anti-spoofing models.

Bo Peng, Daniel Goldstein, Quentin Anthony et al. · harvard

We present Eagle (RWKV-5) and Finch (RWKV-6), sequence models improving upon the RWKV (RWKV-4) architecture. Our architectural design advancements include multi-headed matrix-valued states and a dynamic recurrence mechanism that improve expressivity while maintaining the inference efficiency characteristics of RNNs. We introduce a new multilingual corpus with 1.12 trillion tokens and a fast tokenizer based on greedy matching for enhanced multilinguality. We trained four Eagle models, ranging from 0.46 to 7.5 billion parameters, and two Finch models with 1.6 and 3.1 billion parameters and find that they achieve competitive performance across a wide variety of benchmarks. We release all our models on HuggingFace under the Apache 2.0 license. Models at: https://huggingface.co/RWKV Training code at: https://github.com/RWKV/RWKV-LM Inference code at: https://github.com/RWKV/ChatRWKV Time-parallel training code at: https://github.com/RWKV/RWKV-infctx-trainer

Minghao Yan, Bo Peng, Benjamin Coleman et al.

Large Language Models (LLMs) have emerged as powerful operators for evolutionary search, yet the design of efficient search scaffolds remains ad hoc. While promising, current LLM-in-the-loop systems lack a systematic approach to managing the evolutionary process. We identify three distinct failure modes: Context Pollution, where experiment history biases future candidate generation; Mode Collapse, where agents stagnate in local minima due to poor exploration-exploitation balance; and Weak Collaboration, where rigid crossover strategies fail to leverage parallel search trajectories effectively. We introduce Progress-Aware Consistent Evolution (PACEvolve), a framework designed to robustly govern the agent's context and search dynamics, to address these challenges. PACEvolve combines hierarchical context management (HCM) with pruning to address context pollution; momentum-based backtracking (MBB) to escape local minima; and a self-adaptive sampling policy that unifies backtracking and crossover for dynamic search coordination (CE), allowing agents to balance internal refinement with cross-trajectory collaboration. We demonstrate that PACEvolve provides a systematic path to consistent, long-horizon self-improvement, achieving state-of-the-art results on LLM-SR and KernelBench, while discovering solutions surpassing the record on Modded NanoGPT.

Bo Peng, Xinyuan Chen, Yaohui Wang et al.

Recent works have successfully extended large-scale text-to-image models to the video domain, producing promising results but at a high computational cost and requiring a large amount of video data. In this work, we introduce ConditionVideo, a training-free approach to text-to-video generation based on the provided condition, video, and input text, by leveraging the power of off-the-shelf text-to-image generation methods (e.g., Stable Diffusion). ConditionVideo generates realistic dynamic videos from random noise or given scene videos. Our method explicitly disentangles the motion representation into condition-guided and scenery motion components. To this end, the ConditionVideo model is designed with a UNet branch and a control branch. To improve temporal coherence, we introduce sparse bi-directional spatial-temporal attention (sBiST-Attn). The 3D control network extends the conventional 2D controlnet model, aiming to strengthen conditional generation accuracy by additionally leveraging the bi-directional frames in the temporal domain. Our method exhibits superior performance in terms of frame consistency, clip score, and conditional accuracy, outperforming other compared methods.

Bo Peng, Ruichong Zhang, Daniel Goldstein et al.

We present RWKV-7 "Goose", a new sequence modeling architecture with constant memory usage and constant inference time per token. Despite being trained on dramatically fewer tokens than other top models, our 2.9 billion parameter language model achieves a new 3B SoTA on multilingual tasks and matches the current 3B SoTA on English language downstream performance. RWKV-7 introduces a newly generalized formulation of the delta rule with vector-valued gating and in-context learning rates, as well as a relaxed value replacement rule. We show that RWKV-7 can perform state tracking and recognize all regular languages, while retaining parallelizability of training. This exceeds the capabilities of Transformers under standard complexity conjectures, which are limited to $\mathsf{TC}^0$. To demonstrate RWKV-7's language modeling capability, we also present an extended open source 3.1 trillion token multilingual corpus, and train four RWKV-7 models ranging from 0.19 billion to 2.9 billion parameters on this dataset. To foster openness, reproduction, and adoption, we release our models and dataset component listing at https://huggingface.co/RWKV, and our training and inference code at https://github.com/RWKV/RWKV-LM all under the Apache 2.0 License.

Bo Peng, Yunfan Tao, Haoyu Zhan et al.

We introduce PICA, a novel representation for high-fidelity animatable clothed human avatars with physics-accurate dynamics, even for loose clothing. Previous neural rendering-based representations of animatable clothed humans typically employ a single model to represent both the clothing and the underlying body. While efficient, these approaches often fail to accurately represent complex garment dynamics, leading to incorrect deformations and noticeable rendering artifacts, especially for sliding or loose garments. Furthermore, previous works represent garment dynamics as pose-dependent deformations and facilitate novel pose animations in a data-driven manner. This often results in outcomes that do not faithfully represent the mechanics of motion and are prone to generating artifacts in out-of-distribution poses. To address these issues, we adopt two individual 3D Gaussian Splatting (3DGS) models with different deformation characteristics, modeling the human body and clothing separately. This distinction allows for better handling of their respective motion characteristics. With this representation, we integrate a graph neural network (GNN)-based clothed body physics simulation module to ensure an accurate representation of clothing dynamics. Our method, through its carefully designed features, achieves high-fidelity rendering of clothed human bodies in complex and novel driving poses, significantly outperforming previous methods under the same settings.

Bo Peng, Xinyi Ling, Ziru Chen et al.

With tremendous efforts on developing effective e-commerce models, conventional e-commerce models show limited success in generalist e-commerce modeling, and suffer from unsatisfactory performance on new users and new products - a typical out-of-domain generalization challenge. Meanwhile, large language models (LLMs) demonstrate outstanding performance in generalist modeling and out-of-domain generalizability in many fields. Toward fully unleashing their power for e-commerce, in this paper, we construct ECInstruct, the first open-sourced, large-scale, and high-quality benchmark instruction dataset for e-commerce. Leveraging ECInstruct, we develop eCeLLM, a series of e-commerce LLMs, by instruction-tuning general-purpose LLMs. Our comprehensive experiments and evaluation demonstrate that eCeLLM models substantially outperform baseline models, including the most advanced GPT-4, and the state-of-the-art task-specific models in in-domain evaluation. Moreover, eCeLLM exhibits excellent generalizability to out-of-domain settings, including unseen products and unseen instructions, highlighting its superiority as a generalist e-commerce model. Both the ECInstruct dataset and the eCeLLM models show great potential in empowering versatile and effective LLMs for e-commerce. ECInstruct and eCeLLM models are publicly accessible through https://ninglab.github.io/eCeLLM.

Tingting Zhu, Bo Peng, Jifan Liang et al.

Semantic communications (SC) have been expected to be a new paradigm shifting to catalyze the next generation communication, whose main concerns shift from accurate bit transmission to effective semantic information exchange in communications. However, the previous and widely-used metrics for images are not applicable to evaluate the image semantic similarity in SC. Classical metrics to measure the similarity between two images usually rely on the pixel level or the structural level, such as the PSNR and the MS-SSIM. Straightforwardly using some tailored metrics based on deep-learning methods in CV community, such as the LPIPS, is infeasible for SC. To tackle this, inspired by BERTScore in NLP community, we propose a novel metric for evaluating image semantic similarity, named Vision Transformer Score (ViTScore). We prove theoretically that ViTScore has 3 important properties, including symmetry, boundedness, and normalization, which make ViTScore convenient and intuitive for image measurement. To evaluate the performance of ViTScore, we compare ViTScore with 3 typical metrics (PSNR, MS-SSIM, and LPIPS) through 4 classes of experiments: (i) correlation with BERTScore through evaluation of image caption downstream CV task, (ii) evaluation in classical image communications, (iii) evaluation in image semantic communication systems, and (iv) evaluation in image semantic communication systems with semantic attack. Experimental results demonstrate that ViTScore is robust and efficient in evaluating the semantic similarity of images. Particularly, ViTScore outperforms the other 3 typical metrics in evaluating the image semantic changes by semantic attack, such as image inverse with Generative Adversarial Networks (GANs). This indicates that ViTScore is an effective performance metric when deployed in SC scenarios.

Zheling Meng, Bo Peng, Jing Dong

Watermarking is a tool for actively identifying and attributing the images generated by latent diffusion models. Existing methods face the dilemma of image quality and watermark robustness. Watermarks with superior image quality usually have inferior robustness against attacks such as blurring and JPEG compression, while watermarks with superior robustness usually significantly damage image quality. This dilemma stems from the traditional paradigm where watermarks are injected and detected in pixel space, relying on pixel perturbation for watermark detection and resilience against attacks. In this paper, we highlight that an effective solution to the problem is to both inject and detect watermarks in the latent diffusion space, and propose Latent Watermark with a progressive training strategy. It weakens the direct connection between quality and robustness and thus alleviates their contradiction. We conduct evaluations on two datasets and against 10 watermark attacks. Six metrics measure the image quality and watermark robustness. Results show that compared to the recently proposed methods such as StableSignature, StegaStamp, RoSteALS, LaWa, TreeRing, and DiffuseTrace, LW not only surpasses them in terms of robustness but also offers superior image quality. Our code will be available at https://github.com/RichardSunnyMeng/LatentWatermark.

Bowen Peng, Bo Peng, Jingyuan Xia et al.

Recently, there has been increasing concern about the vulnerability of deep neural network (DNN)-based synthetic aperture radar (SAR) automatic target recognition (ATR) to adversarial attacks, where a DNN could be easily deceived by clean input with imperceptible but aggressive perturbations. This paper studies the synthetic-to-measured (S2M) transfer setting, where an attacker generates adversarial perturbation based solely on synthetic data and transfers it against victim models trained with measured data. Compared with the current measured-to-measured (M2M) transfer setting, our approach does not need direct access to the victim model or the measured SAR data. We also propose the transferability estimation attack (TEA) to uncover the adversarial risks in this more challenging and practical scenario. The TEA makes full use of the limited similarity between the synthetic and measured data pairs for blind estimation and optimization of S2M transferability, leading to feasible surrogate model enhancement without mastering the victim model and data. Comprehensive evaluations based on the publicly available synthetic and measured paired labeled experiment (SAMPLE) dataset demonstrate that the TEA outperforms state-of-the-art methods and can significantly enhance various attack algorithms in computer vision and remote sensing applications. Codes and data are available at https://github.com/scenarri/S2M-TEA.

Jia Zhang, Bo Peng, Xi Wu

Currently, existing efforts in Weakly Supervised Semantic Segmentation (WSSS) based on Convolutional Neural Networks (CNNs) have predominantly focused on enhancing the multi-label classification network stage, with limited attention given to the equally important downstream segmentation network. Furthermore, CNN-based local convolutions lack the ability to model the extensive inter-category dependencies. Therefore, this paper introduces a graph reasoning-based approach to enhance WSSS. The aim is to improve WSSS holistically by simultaneously enhancing both the multi-label classification and segmentation network stages. In the multi-label classification network segment, external knowledge is integrated, coupled with GCNs, to globally reason about inter-class dependencies. This encourages the network to uncover features in non-salient regions of images, thereby refining the completeness of generated pseudo-labels. In the segmentation network segment, the proposed Graph Reasoning Mapping (GRM) module is employed to leverage knowledge obtained from textual databases, facilitating contextual reasoning for class representation within image regions. This GRM module enhances feature representation in high-level semantics of the segmentation network's local convolutions, while dynamically learning semantic coherence for individual samples. Using solely image-level supervision, we have achieved state-of-the-art performance in WSSS on the PASCAL VOC 2012 and MS-COCO datasets. Extensive experimentation on both the multi-label classification and segmentation network stages underscores the effectiveness of the proposed graph reasoning approach for advancing WSSS.

Wenbing Zhu, Chengjie Wang, Bin-Bin Gao et al.

Industrial Anomaly Detection (IAD) is critical for enhancing operational safety, ensuring product quality, and optimizing manufacturing efficiency across global industries. However, the IAD algorithms are severely constrained by the limitations of existing public benchmarks. Current datasets exhibit restricted category diversity and insufficient scale, frequently resulting in metric saturation and limited model transferability to real-world scenarios. To address this gap, we introduce Real-IAD Variety, the largest and most diverse IAD benchmark, comprising 198,960 high-resolution images across 160 distinct object categories. Its diversity is ensured through comprehensive coverage of 28 industries, 24 material types, and 22 color variations. Our comprehensive experimental analysis validates the benchmark's substantial challenge: state-of-the-art multi-class unsupervised anomaly detection methods experience significant performance degradation when scaled from 30 to 160 categories. Crucially, we demonstrate that vision-language models exhibit remarkable robustness to category scale-up, with minimal performance variation across different category counts, significantly enhancing generalization capabilities in diverse industrial contexts. The unprecedented scale and complexity of Real-IAD Variety position it as an essential resource for training and evaluating next-generation foundation models for anomaly detection. By providing this comprehensive benchmark with rigorous evaluation protocols across multi-class unsupervised, multi-view, and zero-/few-shot settings, we aim to accelerate research beyond domain-specific constraints, enabling the development of scalable, general-purpose anomaly detection systems. Real-IAD Variety will be made publicly available to facilitate innovation in this critical field.

Zheling Meng, Bo Peng, Xiaochuan Jin et al.

Text-to-image diffusion models have been demonstrated with undesired generation due to unfiltered large-scale training data, such as sexual images and copyrights, necessitating the erasure of undesired concepts. Most existing methods focus on modifying the generation probabilities conditioned on the texts containing target concepts. However, they fail to guarantee the desired generation of texts unseen in the training phase, especially for the adversarial texts from malicious attacks. In this paper, we analyze the erasure task and point out that existing methods cannot guarantee the minimization of the total probabilities of undesired generation. To tackle this problem, we propose Dark Miner. It entails a recurring three-stage process that comprises mining, verifying, and circumventing. This method greedily mines embeddings with maximum generation probabilities of target concepts and more effectively reduces their generation. In the experiments, we evaluate its performance on the inappropriateness, object, and style concepts. Compared with the previous methods, our method achieves better erasure and defense results, especially under multiple adversarial attacks, while preserving the native generation capability of the models. Our code will be available on GitHub.

Shunian Xiang, Patrick J. Lawrence, Bo Peng et al.

Recently, drug repurposing has emerged as an effective and resource-efficient paradigm for AD drug discovery. Among various methods for drug repurposing, network-based methods have shown promising results as they are capable of leveraging complex networks that integrate multiple interaction types, such as protein-protein interactions, to more effectively identify candidate drugs. However, existing approaches typically assume paths of the same length in the network have equal importance in identifying the therapeutic effect of drugs. Other domains have found that same length paths do not necessarily have the same importance. Thus, relying on this assumption may be deleterious to drug repurposing attempts. In this work, we propose MPI (Modeling Path Importance), a novel network-based method for AD drug repurposing. MPI is unique in that it prioritizes important paths via learned node embeddings, which can effectively capture a network's rich structural information. Thus, leveraging learned embeddings allows MPI to effectively differentiate the importance among paths. We evaluate MPI against a commonly used baseline method that identifies anti-AD drug candidates primarily based on the shortest paths between drugs and AD in the network. We observe that among the top-50 ranked drugs, MPI prioritizes 20.0% more drugs with anti-AD evidence compared to the baseline. Finally, Cox proportional-hazard models produced from insurance claims data aid us in identifying the use of etodolac, nicotine, and BBB-crossing ACE-INHs as having a reduced risk of AD, suggesting such drugs may be viable candidates for repurposing and should be explored further in future studies.

Xin Yang, Dongliang Fan, Yunteng Ma et al.

Cardiovascular diseases account for around 17.9 million deaths per year globally, the treatment of which is challenging considering the confined space and complex topology of the vascular network and high risks during operations. Robots, although promising, still face the dilemma of possessing versatility or maneuverability after decades of development. Inspired by nematodes, the parasites living, feeding, and moving in the human body's vascular system, this work develops a transformable slender magnetic microrobot. Based on the experiments and analyses, we optimize the fabrication and geometry of the robot and finally create a slender prototype with an aspect ratio larger than 100 (smaller than 200 microns in diameter and longer than 20 mm in length), which possesses uniformly distributed magnetic beads on the body of an ultrathin polymer string and a big bead on the head. This prototype shows great flexibility (largest curvature 0.904 mm-1) and locomotion capability (the maximum speed: 125 mm/s). Moreover, the nematode-inspired robot can pass through sharp turns with a radius of 0.84 mm and holes distributed in three-dimensional (3D) space. We also display the potential application in interventional surgery of the microrobot by navigating it through a narrow blood vessel mold to wrap and transport a drug (95 times heavier than the robot) by deforming the robot's slender body and releasing the drug to the aim position finally. Moreover, the robot also demonstrates the possible applications in embolization by transforming and winding itself into an aneurysms phantom and exhibits its outstanding injectability by being successfully withdrawn and injected through a medical needle (diameter: 1.2 mm) of a syringe.

Peng Huang, Shu Hu, Bo Peng et al.

MedSAM, a medical foundation model derived from the SAM architecture, has demonstrated notable success across diverse medical domains. However, its clinical application faces two major challenges: the dependency on labor-intensive manual prompt generation, which imposes a significant burden on clinicians, and the absence of semantic labeling in the generated segmentation masks for organs or lesions, limiting its practicality for non-expert users. To address these limitations, we propose AutoMedSAM, an end-to-end framework derived from SAM, designed to enhance usability and segmentation performance. AutoMedSAM retains MedSAM's image encoder and mask decoder structure while introducing a novel diffusion-based class prompt encoder. The diffusion-based encoder employs a dual-decoder structure to collaboratively generate prompt embeddings guided by sparse and dense prompt definitions. These embeddings enhance the model's ability to understand and process clinical imagery autonomously. With this encoder, AutoMedSAM leverages class prompts to embed semantic information into the model's predictions, transforming MedSAM's semi-automated pipeline into a fully automated workflow. Furthermore, AutoMedSAM employs an uncertainty-aware joint optimization strategy during training to effectively inherit MedSAM's pre-trained knowledge while improving generalization by integrating multiple loss functions. Experimental results across diverse datasets demonstrate that AutoMedSAM achieves superior performance while broadening its applicability to both clinical settings and non-expert users. Code is available at https://github.com/HP-ML/AutoPromptMedSAM.git.

Hao Chen, Jiafu Wu, Ying Jin et al.

Recently, methods like Zero-1-2-3 have focused on single-view based 3D reconstruction and have achieved remarkable success. However, their predictions for unseen areas heavily rely on the inductive bias of large-scale pretrained diffusion models. Although subsequent work, such as DreamComposer, attempts to make predictions more controllable by incorporating additional views, the results remain unrealistic due to feature entanglement in the vanilla latent space, including factors such as lighting, material, and structure. To address these issues, we introduce the Visual Isotropy 3D Reconstruction Model (VI3DRM), a diffusion-based sparse views 3D reconstruction model that operates within an ID consistent and perspective-disentangled 3D latent space. By facilitating the disentanglement of semantic information, color, material properties and lighting, VI3DRM is capable of generating highly realistic images that are indistinguishable from real photographs. By leveraging both real and synthesized images, our approach enables the accurate construction of pointmaps, ultimately producing finely textured meshes or point clouds. On the NVS task, tested on the GSO dataset, VI3DRM significantly outperforms state-of-the-art method DreamComposer, achieving a PSNR of 38.61, an SSIM of 0.929, and an LPIPS of 0.027. Code will be made available upon publication.

Chuanyang Zheng, Jiankai Sun, Yihang Gao et al.

The placement of normalization layers, specifically Pre-Norm and Post-Norm, remains an open question in Transformer architecture design. In this work, we rethink these approaches through the lens of manifold optimization, interpreting the outputs of the Feed-Forward Network (FFN) and attention layers as update directions in optimization. Building on this perspective, we introduce GeoNorm, a novel method that replaces standard normalization with geodesic updates on the manifold. Furthermore, analogous to learning rate schedules, we propose a layer-wise update decay for the FFN and attention components. Comprehensive experiments demonstrate that GeoNorm consistently outperforms existing normalization methods in Transformer models. Crucially, GeoNorm can be seamlessly integrated into standard Transformer architectures, achieving performance improvements with negligible additional computational cost.

Yuanjie Lyu, Tong Xu, Zihan Niu et al.

The prosperity of social media platforms has raised the urgent demand for semantic-rich services, e.g., event and storyline attribution. However, most existing research focuses on clip-level event understanding, primarily through basic captioning tasks, without analyzing the causes of events across an entire movie. This is a significant challenge, as even advanced multimodal large language models (MLLMs) struggle with extensive multimodal information due to limited context length. To address this issue, we propose a Two-Stage Prefix-Enhanced MLLM (TSPE) approach for event attribution, i.e., connecting associated events with their causal semantics, in movie videos. In the local stage, we introduce an interaction-aware prefix that guides the model to focus on the relevant multimodal information within a single clip, briefly summarizing the single event. Correspondingly, in the global stage, we strengthen the connections between associated events using an inferential knowledge graph, and design an event-aware prefix that directs the model to focus on associated events rather than all preceding clips, resulting in accurate event attribution. Comprehensive evaluations of two real-world datasets demonstrate that our framework outperforms state-of-the-art methods.

Bo Peng, Pi Bu, Keyu Pan et al.

Recent advances in vision-language models (VLMs) have shown promise for human-level embodied intelligence. However, existing benchmarks for VLM-driven embodied agents often rely on high-level commands or discretized action spaces, which are non-native settings that differ markedly from real-world control. In addition, current benchmarks focus primarily on high-level tasks and lack joint evaluation and analysis at both low and high levels. To address these limitations, we present NativeEmbodied, a challenging benchmark for VLM-driven embodied agents that uses a unified, native low-level action space. Built on diverse simulated scenes, NativeEmbodied includes three representative high-level tasks in complex scenarios to evaluate overall performance. For more detailed analysis, we further decouple the skills required by complex tasks and construct four types of low-level tasks, each targeting a fundamental embodied skill. This joint evaluation across task and skill granularities enables fine-grained assessment of embodied agents. Experiments with state-of-the-art VLMs reveal clear deficiencies in several fundamental embodied skills, and further analysis shows that these bottlenecks significantly limit performance on high-level tasks. NativeEmbodied highlights key challenges for current VLM-driven embodied agents and provides insights to guide future research.

Runyuan He, Qiuyang Mang, Shang Zhou et al.

Many real-world coding challenges are open-ended and admit no known optimal solution. Yet, recent progress in LLM coding has focused on well-defined tasks such as feature implementation, bug fixing, and competitive programming. Open-ended coding remains a weak spot for LLMs, largely because open-ended training problems are scarce and expensive to construct. Our goal is to synthesize open-ended coding problems at scale to train stronger LLM coders. We introduce FrontierSmith, an automated system for iteratively evolving open-ended problems from existing closed-ended coding tasks. Starting from competitive programming problems, FrontierSmith generates candidate open-ended variants by changing the problems'goals, restricting outputs, and generalizing inputs. It then uses a quantitative idea divergence metric to select problems that elicit genuinely diverse approaches from different solvers. Agents then generate test cases and verifiers for the surviving candidates. On two open-ended coding benchmarks, training on our synthesized data yields substantial gains over the base models: Qwen3.5-9B improves by +8.82 score on FrontierCS and +306.36 (Elo-rating-based performance) on ALE-bench; Qwen3.5-27B improves by +12.12 and +309.12, respectively. The synthesized problems also make agents take more turns and use more tokens, similar to human-curated ones, suggesting that closed-ended seeds can be a practical starting point for long-horizon coding data.

Bo Peng, Yuanwei Hu, Bo Liu et al.

Spectral clustering is known as a powerful technique in unsupervised data analysis. The vast majority of approaches to spectral clustering are driven by a single modality, leaving the rich information in multi-modal representations untapped. Inspired by the recent success of vision-language pre-training, this paper enriches the landscape of spectral clustering from a single-modal to a multi-modal regime. Particularly, we propose Neural Tangent Kernel Spectral Clustering that leverages cross-modal alignment in pre-trained vision-language models. By anchoring the neural tangent kernel with positive nouns, i.e., those semantically close to the images of interest, we arrive at formulating the affinity between images as a coupling of their visual proximity and semantic overlap. We show that this formulation amplifies within-cluster connections while suppressing spurious ones across clusters, hence encouraging block-diagonal structures. In addition, we present a regularized affinity diffusion mechanism that adaptively ensembles affinity matrices induced by different prompts. Extensive experiments on \textbf{16} benchmarks -- including classical, large-scale, fine-grained and domain-shifted datasets -- manifest that our method consistently outperforms the state-of-the-art by a large margin.