Xiangyu Yue

Peng Gao, Jiaming Han, Renrui Zhang et al. · berkeley, stanford

How to efficiently transform large language models (LLMs) into instruction followers is recently a popular research direction, while training LLM for multi-modal reasoning remains less explored. Although the recent LLaMA-Adapter demonstrates the potential to handle visual inputs with LLMs, it still cannot generalize well to open-ended visual instructions and lags behind GPT-4. In this paper, we present LLaMA-Adapter V2, a parameter-efficient visual instruction model. Specifically, we first augment LLaMA-Adapter by unlocking more learnable parameters (e.g., norm, bias and scale), which distribute the instruction-following ability across the entire LLaMA model besides adapters. Secondly, we propose an early fusion strategy to feed visual tokens only into the early LLM layers, contributing to better visual knowledge incorporation. Thirdly, a joint training paradigm of image-text pairs and instruction-following data is introduced by optimizing disjoint groups of learnable parameters. This strategy effectively alleviates the interference between the two tasks of image-text alignment and instruction following and achieves strong multi-modal reasoning with only a small-scale image-text and instruction dataset. During inference, we incorporate additional expert models (e.g. captioning/OCR systems) into LLaMA-Adapter to further enhance its image understanding capability without incurring training costs. Compared to the original LLaMA-Adapter, our LLaMA-Adapter V2 can perform open-ended multi-modal instructions by merely introducing 14M parameters over LLaMA. The newly designed framework also exhibits stronger language-only instruction-following capabilities and even excels in chat interactions. Our code and models are available at https://github.com/ZrrSkywalker/LLaMA-Adapter.

Yiyuan Zhang, Kaixiong Gong, Kaipeng Zhang et al. · berkeley

Multimodal learning aims to build models that can process and relate information from multiple modalities. Despite years of development in this field, it still remains challenging to design a unified network for processing various modalities ($\textit{e.g.}$ natural language, 2D images, 3D point clouds, audio, video, time series, tabular data) due to the inherent gaps among them. In this work, we propose a framework, named Meta-Transformer, that leverages a $\textbf{frozen}$ encoder to perform multimodal perception without any paired multimodal training data. In Meta-Transformer, the raw input data from various modalities are mapped into a shared token space, allowing a subsequent encoder with frozen parameters to extract high-level semantic features of the input data. Composed of three main components: a unified data tokenizer, a modality-shared encoder, and task-specific heads for downstream tasks, Meta-Transformer is the first framework to perform unified learning across 12 modalities with unpaired data. Experiments on different benchmarks reveal that Meta-Transformer can handle a wide range of tasks including fundamental perception (text, image, point cloud, audio, video), practical application (X-Ray, infrared, hyperspectral, and IMU), and data mining (graph, tabular, and time-series). Meta-Transformer indicates a promising future for developing unified multimodal intelligence with transformers. Code will be available at https://github.com/invictus717/MetaTransformer

Jiaming Han, Renrui Zhang, Wenqi Shao et al. · berkeley

We present ImageBind-LLM, a multi-modality instruction tuning method of large language models (LLMs) via ImageBind. Existing works mainly focus on language and image instruction tuning, different from which, our ImageBind-LLM can respond to multi-modality conditions, including audio, 3D point clouds, video, and their embedding-space arithmetic by only image-text alignment training. During training, we adopt a learnable bind network to align the embedding space between LLaMA and ImageBind's image encoder. Then, the image features transformed by the bind network are added to word tokens of all layers in LLaMA, which progressively injects visual instructions via an attention-free and zero-initialized gating mechanism. Aided by the joint embedding of ImageBind, the simple image-text training enables our model to exhibit superior multi-modality instruction-following capabilities. During inference, the multi-modality inputs are fed into the corresponding ImageBind encoders, and processed by a proposed visual cache model for further cross-modal embedding enhancement. The training-free cache model retrieves from three million image features extracted by ImageBind, which effectively mitigates the training-inference modality discrepancy. Notably, with our approach, ImageBind-LLM can respond to instructions of diverse modalities and demonstrate significant language generation quality. Code is released at https://github.com/OpenGVLab/LLaMA-Adapter.

Yiyuan Zhang, Kaixiong Gong, Xiaohan Ding et al. · berkeley

We propose $\textbf{UniDG}$, a novel and $\textbf{Uni}$fied framework for $\textbf{D}$omain $\textbf{G}$eneralization that is capable of significantly enhancing the out-of-distribution generalization performance of foundation models regardless of their architectures. The core idea of UniDG is to finetune models during the inference stage, which saves the cost of iterative training. Specifically, we encourage models to learn the distribution of test data in an unsupervised manner and impose a penalty regarding the updating step of model parameters. The penalty term can effectively reduce the catastrophic forgetting issue as we would like to maximally preserve the valuable knowledge in the original model. Empirically, across 12 visual backbones, including CNN-, MLP-, and Transformer-based models, ranging from 1.89M to 303M parameters, UniDG shows an average accuracy improvement of +5.4% on DomainBed. These performance results demonstrate the superiority and versatility of UniDG. The code is publicly available at https://github.com/invictus717/UniDG

Zangwei Zheng, Mingyuan Ma, Kai Wang et al. · berkeley

Continual learning (CL) can help pre-trained vision-language models efficiently adapt to new or under-trained data distributions without re-training. Nevertheless, during the continual training of the Contrastive Language-Image Pre-training (CLIP) model, we observe that the model's zero-shot transfer ability significantly degrades due to catastrophic forgetting. Existing CL methods can mitigate forgetting by replaying previous data. However, since the CLIP dataset is private, replay methods cannot access the pre-training dataset. In addition, replaying data of previously learned downstream tasks can enhance their performance but comes at the cost of sacrificing zero-shot performance. To address this challenge, we propose a novel method ZSCL to prevent zero-shot transfer degradation in the continual learning of vision-language models in both feature and parameter space. In the feature space, a reference dataset is introduced for distillation between the current and initial models. The reference dataset should have semantic diversity but no need to be labeled, seen in pre-training, or matched image-text pairs. In parameter space, we prevent a large parameter shift by averaging weights during the training. We propose a more challenging Multi-domain Task Incremental Learning (MTIL) benchmark to evaluate different methods, where tasks are from various domains instead of class-separated in a single dataset. Our method outperforms other methods in the traditional class-incremental learning setting and the MTIL by 9.7% average score. Our code locates at https://github.com/Thunderbeee/ZSCL.

Min Liu, Alberto Sangiovanni-Vincentelli, Xiangyu Yue · berkeley

Deep neural networks (DNNs) are vulnerable to backdoor attack, which does not affect the network's performance on clean data but would manipulate the network behavior once a trigger pattern is added. Existing defense methods have greatly reduced attack success rate, but their prediction accuracy on clean data still lags behind a clean model by a large margin. Inspired by the stealthiness and effectiveness of backdoor attack, we propose a simple but highly effective defense framework which injects non-adversarial backdoors targeting poisoned samples. Following the general steps in backdoor attack, we detect a small set of suspected samples and then apply a poisoning strategy to them. The non-adversarial backdoor, once triggered, suppresses the attacker's backdoor on poisoned data, but has limited influence on clean data. The defense can be carried out during data preprocessing, without any modification to the standard end-to-end training pipeline. We conduct extensive experiments on multiple benchmarks with different architectures and representative attacks. Results demonstrate that our method achieves state-of-the-art defense effectiveness with by far the lowest performance drop on clean data. Considering the surprising defense ability displayed by our framework, we call for more attention to utilizing backdoor for backdoor defense. Code is available at https://github.com/minliu01/non-adversarial_backdoor.

Xiaohan Ding, Yiyuan Zhang, Yixiao Ge et al. · tencent-ai

Large-kernel convolutional neural networks (ConvNets) have recently received extensive research attention, but two unresolved and critical issues demand further investigation. 1) The architectures of existing large-kernel ConvNets largely follow the design principles of conventional ConvNets or transformers, while the architectural design for large-kernel ConvNets remains under-addressed. 2) As transformers have dominated multiple modalities, it remains to be investigated whether ConvNets also have a strong universal perception ability in domains beyond vision. In this paper, we contribute from two aspects. 1) We propose four architectural guidelines for designing large-kernel ConvNets, the core of which is to exploit the essential characteristics of large kernels that distinguish them from small kernels - they can see wide without going deep. Following such guidelines, our proposed large-kernel ConvNet shows leading performance in image recognition (ImageNet accuracy of 88.0%, ADE20K mIoU of 55.6%, and COCO box AP of 56.4%), demonstrating better performance and higher speed than the recent powerful competitors. 2) We discover large kernels are the key to unlocking the exceptional performance of ConvNets in domains where they were originally not proficient. With certain modality-related preprocessing approaches, the proposed model achieves state-of-the-art performance on time-series forecasting and audio recognition tasks even without modality-specific customization to the architecture. All the code and models are publicly available on GitHub and Huggingface.

Haodi He, Yuhui Yuan, Xiangyu Yue et al. · berkeley

The segmentation task has traditionally been formulated as a complete-label pixel classification task to predict a class for each pixel from a fixed number of predefined semantic categories shared by all images or videos. Yet, following this formulation, standard architectures will inevitably encounter various challenges under more realistic settings where the scope of categories scales up (e.g., beyond the level of 1k). On the other hand, in a typical image or video, only a few categories, i.e., a small subset of the complete label are present. Motivated by this intuition, in this paper, we propose to decompose segmentation into two sub-problems: (i) image-level or video-level multi-label classification and (ii) pixel-level rank-adaptive selected-label classification. Given an input image or video, our framework first conducts multi-label classification over the complete label, then sorts the complete label and selects a small subset according to their class confidence scores. We then use a rank-adaptive pixel classifier to perform the pixel-wise classification over only the selected labels, which uses a set of rank-oriented learnable temperature parameters to adjust the pixel classifications scores. Our approach is conceptually general and can be used to improve various existing segmentation frameworks by simply using a lightweight multi-label classification head and rank-adaptive pixel classifier. We demonstrate the effectiveness of our framework with competitive experimental results across four tasks, including image semantic segmentation, image panoptic segmentation, video instance segmentation, and video semantic segmentation. Especially, with our RankSeg, Mask2Former gains +0.8%/+0.7%/+0.7% on ADE20K panoptic segmentation/YouTubeVIS 2019 video instance segmentation/VSPW video semantic segmentation benchmarks respectively.

Zangwei Zheng, Xiangyu Yue, Kai Wang et al. · berkeley

Though vision transformers (ViTs) have exhibited impressive ability for representation learning, we empirically find that they cannot generalize well to unseen domains with previous domain generalization algorithms. In this paper, we propose a novel approach DoPrompt based on prompt learning to embed the knowledge of source domains in domain prompts for target domain prediction. Specifically, domain prompts are prepended before ViT input tokens from the corresponding source domain. Each domain prompt learns domain-specific knowledge efficiently since it is optimized only for one domain. Meanwhile, we train a prompt adapter to produce a suitable prompt for each input image based on the learned source domain prompts. At test time, the adapted prompt generated by the prompt adapter can exploit the similarity between the feature of the out-of-domain image and source domains to properly integrate the source domain knowledge. Extensive experiments are conducted on four benchmark datasets. Our approach achieves 1.4% improvements in the averaged accuracy, which is 3.5 times the improvement of the state-of-the-art algorithm with a ViT backbone.

Zhanhui Zhou, Jie Liu, Jing Shao et al.

A single language model, even when aligned with labelers through reinforcement learning from human feedback (RLHF), may not suit all human preferences. Recent approaches therefore prefer customization, gathering multi-dimensional feedback, and creating distinct reward models for each dimension. Different language models are then optimized for various preferences using multi-objective RLHF (MORLHF) with varying reward weights. However, RL fine-tuning is unstable and resource-heavy, especially with diverse and usually conflicting objectives. In this paper, we present Multi-Objective Direct Preference Optimization (MODPO), an RL-free extension of Direct Preference Optimization (DPO) for multiple alignment objectives. Essentially, MODPO folds language modeling directly into reward modeling, training language models as implicit collective reward models that combine all objectives with specific weights. MODPO theoretically yields the same optimal solutions as MORLHF but is practically more stable and efficient. Empirical results in safety alignment and long-form question answering show that MODPO matches or outperforms existing methods, producing a Pareto front of language models catering to diverse preferences with three times less computational resources compared to MORLHF. Code is available at https://github.com/ZHZisZZ/modpo.

Changqi Wang, Haoyu Xie, Yuhui Yuan et al. · berkeley

Semi-Supervised Semantic Segmentation (S4) aims to train a segmentation model with limited labeled images and a substantial volume of unlabeled images. To improve the robustness of representations, powerful methods introduce a pixel-wise contrastive learning approach in latent space (i.e., representation space) that aggregates the representations to their prototypes in a fully supervised manner. However, previous contrastive-based S4 methods merely rely on the supervision from the model's output (logits) in logit space during unlabeled training. In contrast, we utilize the outputs in both logit space and representation space to obtain supervision in a collaborative way. The supervision from two spaces plays two roles: 1) reduces the risk of over-fitting to incorrect semantic information in logits with the help of representations; 2) enhances the knowledge exchange between the two spaces. Furthermore, unlike previous approaches, we use the similarity between representations and prototypes as a new indicator to tilt training those under-performing representations and achieve a more efficient contrastive learning process. Results on two public benchmarks demonstrate the competitive performance of our method compared with state-of-the-art methods.

Yan Ma, Weicong Liang, Bohan Chen et al. · berkeley

Motivated by the remarkable achievements of DETR-based approaches on COCO object detection and segmentation benchmarks, recent endeavors have been directed towards elevating their performance through self-supervised pre-training of Transformers while preserving a frozen backbone. Noteworthy advancements in accuracy have been documented in certain studies. Our investigation delved deeply into a representative approach, DETReg, and its performance assessment in the context of emerging models like $\mathcal{H}$-Deformable-DETR. Regrettably, DETReg proves inadequate in enhancing the performance of robust DETR-based models under full data conditions. To dissect the underlying causes, we conduct extensive experiments on COCO and PASCAL VOC probing elements such as the selection of pre-training datasets and strategies for pre-training target generation. By contrast, we employ an optimized approach named Simple Self-training which leads to marked enhancements through the combination of an improved box predictor and the Objects$365$ benchmark. The culmination of these endeavors results in a remarkable AP score of $59.3\%$ on the COCO val set, outperforming $\mathcal{H}$-Deformable-DETR + Swin-L without pre-training by $1.4\%$. Moreover, a series of synthetic pre-training datasets, generated by merging contemporary image-to-text(LLaVA) and text-to-image (SDXL) models, significantly amplifies object detection capabilities.

Yilei Jiang, Jinyuan Hu, Qianyin Xiao et al.

Game development sits at the intersection of creative design and intricate software engineering, demanding the joint orchestration of game engines, real-time loops, and tightly coupled state across many files. While Large Language Models (LLMs) and code agents now solve isolated programming tasks with ease, they consistently stumble when asked to produce a fully playable game from a high-level design, collapsing under cross-file inconsistencies, broken scene wiring, and logical incoherence. We bridge this gap with OpenGame, the first open-source agentic framework explicitly designed for end-to-end web game creation. At its core lies Game Skill, a reusable, evolving capability composed of a Template Skill that grows a library of project skeletons from experience and a Debug Skill that maintains a living protocol of verified fixes - together enabling the agent to scaffold stable architectures and systematically repair integration errors rather than patch isolated syntax bugs. Powering this framework is GameCoder-27B, a code LLM specialized for game engine mastery through a three-stage pipeline of continual pre-training, supervised fine-tuning, and execution-grounded reinforcement learning. Since verifying interactive playability is fundamentally harder than checking static code, we further introduce OpenGame-Bench, an evaluation pipeline that scores agentic game generation along Build Health, Visual Usability, and Intent Alignment via headless browser execution and VLM judging. Across 150 diverse game prompts, OpenGame establishes a new state-of-the-art. We hope OpenGame pushes code agents beyond discrete software engineering problems and toward building complex, interactive real-world applications. Our framework will be fully open-sourced.

Wenze Liu, Zixuan Ye, Hao Lu et al.

The labelling difficulty has been a longstanding problem in deep image matting. To escape from fine labels, this work explores using rough annotations such as trimaps coarsely indicating the foreground/background as supervision. We present that the cooperation between learned semantics from indicated known regions and proper assumed matting rules can help infer alpha values at transition areas. Inspired by the nonlocal principle in traditional image matting, we build a directional distance consistency loss (DDC loss) at each pixel neighborhood to constrain the alpha values conditioned on the input image. DDC loss forces the distance of similar pairs on the alpha matte and on its corresponding image to be consistent. In this way, the alpha values can be propagated from learned known regions to unknown transition areas. With only images and trimaps, a matting model can be trained under the supervision of a known loss and the proposed DDC loss. Experiments on AM-2K and P3M-10K dataset show that our paradigm achieves comparable performance with the fine-label-supervised baseline, while sometimes offers even more satisfying results than human-labelled ground truth. Code is available at \url{https://github.com/poppuppy/alpha-free-matting}.

Dongyu Luo, Jianyu Wu, Jing Wang et al.

We showcase the plain diffusion models with Transformers are effective predictors of fluid dynamics under various working conditions, e.g., Darcy flow and high Reynolds number. Unlike traditional fluid dynamical solvers that depend on complex architectures to extract intricate correlations and learn underlying physical states, our approach formulates the prediction of flow dynamics as the image translation problem and accordingly leverage the plain diffusion model to tackle the problem. This reduction in model design complexity does not compromise its ability to capture complex physical states and geometric features of fluid dynamical equations, leading to high-precision solutions. In preliminary tests on various fluid-related benchmarks, our DiffFluid achieves consistent state-of-the-art performance, particularly in solving the Navier-Stokes equations in fluid dynamics, with a relative precision improvement of +44.8%. In addition, we achieved relative improvements of +14.0% and +11.3% in the Darcy flow equation and the airfoil problem with Euler's equation, respectively. Code will be released at https://github.com/DongyuLUO/DiffFluid upon acceptance.

Zhenyu Wu, Jingjing Xie, Zehao Li et al.

With VLM-powered computer-using agents (CUAs) becoming increasingly capable at graphical user interface (GUI) navigation and manipulation, reliable step-level decision-making has emerged as a key bottleneck for real-world deployment. In long-horizon workflows, errors accumulate quickly and irreversible actions can cause unintended consequences, motivating critic models that assess each action before execution. While critic models offer a promising solution, their effectiveness is hindered by the lack of diverse, high-quality GUI feedback data and public critic benchmarks for step-level evaluation in computer use. To bridge these gaps, we introduce OS-Oracle that makes three core contributions: (1) a scalable data pipeline for synthesizing cross-platform GUI critic data; (2) a two-stage training paradigm combining supervised fine-tuning (SFT) and consistency-preserving group relative policy optimization (CP-GRPO); (3) OS-Critic Bench, a holistic benchmark for evaluating critic model performance across Mobile, Web, and Desktop platforms. Leveraging this framework, we curate a high-quality dataset containing 310k critic samples. The resulting critic model, OS-Oracle-7B, achieves state-of-the-art performance among open-source VLMs on OS-Critic Bench, and surpasses proprietary models on the mobile domain. Furthermore, when serving as a pre-critic, OS-Oracle-7B improves the performance of native GUI agents such as UI-TARS-1.5-7B in OSWorld and AndroidWorld environments. The code is open-sourced at https://github.com/numbmelon/OS-Oracle.

Yiliu Yang, Yilei Jiang, Qunzhong Wang et al.

Safety risks arise as large language model-based agents solve complex tasks with tools, multi-step plans, and inter-agent messages. However, deployer-written policies in natural language are ambiguous and context dependent, so they map poorly to machine-checkable rules, and runtime enforcement is unreliable. Expressing safety policies as sequents, we propose \textsc{QuadSentinel}, a four-agent guard (state tracker, policy verifier, threat watcher, and referee) that compiles these policies into machine-checkable rules built from predicates over observable state and enforces them online. Referee logic plus an efficient top-$k$ predicate updater keeps costs low by prioritizing checks and resolving conflicts hierarchically. Measured on ST-WebAgentBench (ICML CUA~'25) and AgentHarm (ICLR~'25), \textsc{QuadSentinel} improves guardrail accuracy and rule recall while reducing false positives. Against single-agent baselines such as ShieldAgent (ICML~'25), it yields better overall safety control. Near-term deployments can adopt this pattern without modifying core agents by keeping policies separate and machine-checkable. Our code will be made publicly available at https://github.com/yyiliu/QuadSentinel.

Jiaming Han, Kaixiong Gong, Yiyuan Zhang et al.

Multimodal large language models (MLLMs) have gained significant attention due to their strong multimodal understanding capability. However, existing works rely heavily on modality-specific encoders, which usually differ in architecture and are limited to common modalities. In this paper, we present OneLLM, an MLLM that aligns eight modalities to language using a unified framework. We achieve this through a unified multimodal encoder and a progressive multimodal alignment pipeline. In detail, we first train an image projection module to connect a vision encoder with LLM. Then, we build a universal projection module (UPM) by mixing multiple image projection modules and dynamic routing. Finally, we progressively align more modalities to LLM with the UPM. To fully leverage the potential of OneLLM in following instructions, we also curated a comprehensive multimodal instruction dataset, including 2M items from image, audio, video, point cloud, depth/normal map, IMU and fMRI brain activity. OneLLM is evaluated on 25 diverse benchmarks, encompassing tasks such as multimodal captioning, question answering and reasoning, where it delivers excellent performance. Code, data, model and online demo are available at https://github.com/csuhan/OneLLM

Kaituo Feng, Kaixiong Gong, Bohao Li et al.

Inspired by DeepSeek-R1's success in eliciting reasoning abilities through rule-based reinforcement learning (RL), we introduce Video-R1 as the first attempt to systematically explore the R1 paradigm for incentivizing video reasoning within multimodal large language models (MLLMs). However, directly applying RL training with the GRPO algorithm to video reasoning presents two primary challenges: (i) a lack of temporal modeling for video reasoning, and (ii) the scarcity of high-quality video-reasoning data. To address these issues, we first propose the T-GRPO algorithm, which encourages models to utilize temporal information in videos for reasoning. Additionally, instead of relying solely on video data, we incorporate high-quality image-reasoning data into the training process. We have constructed two datasets: Video-R1-CoT-165k for SFT cold start and Video-R1-260k for RL training, both comprising image and video data. Experimental results demonstrate that Video-R1 achieves significant improvements on video reasoning benchmarks such as VideoMMMU and VSI-Bench, as well as on general video benchmarks including MVBench and TempCompass, etc. Notably, Video-R1-7B attains a 37.1% accuracy on video spatial reasoning benchmark VSI-bench, surpassing the commercial proprietary model GPT-4o. All code, models, and data are released in: https://github.com/tulerfeng/Video-R1.

Ruize Ma, Yilei Jiang, Shilin Zhang et al.

Multimodal Automated Program Repair (MAPR) extends traditional program repair by requiring models to jointly reason over source code, textual issue descriptions, and visual artifacts such as GUI screenshots. While recent LLM-based repair systems have shown promising results, existing approaches face several limitations: rigid workflow pipelines restrict exploration during debugging, visual reasoning is often performed over full-page screenshots without localized grounding, and failed repair attempts are rarely transformed into reusable knowledge. To address these challenges, we propose FailureMem, a multimodal repair framework that integrates three key mechanisms: a hybrid workflow-agent architecture that balances structured localization with flexible reasoning, active perception tools that enable region-level visual grounding, and a Failure Memory Bank that converts past repair attempts into reusable guidance. Experiments on SWE-bench Multimodal demonstrate FailureMem improves the resolved rate over GUIRepair by 3.7%.

Zhixin Zhang, Yiyuan Zhang, Xiaohan Ding et al.

The construction of online vectorized High-Definition (HD) maps is critical for downstream prediction and planning. Recent efforts have built strong baselines for this task, however, shapes and relations of instances in urban road systems are still under-explored, such as parallelism, perpendicular, or rectangle-shape. In our work, we propose GeMap ($\textbf{Ge}$ometry $\textbf{Map}$), which end-to-end learns Euclidean shapes and relations of map instances beyond basic perception. Specifically, we design a geometric loss based on angle and distance clues, which is robust to rigid transformations. We also decouple self-attention to independently handle Euclidean shapes and relations. Our method achieves new state-of-the-art performance on the NuScenes and Argoverse 2 datasets. Remarkably, it reaches a 71.8% mAP on the large-scale Argoverse 2 dataset, outperforming MapTR V2 by +4.4% and surpassing the 70% mAP threshold for the first time. Code is available at https://github.com/cnzzx/GeMap.

Sihan Yang, Runsen Xu, Yiman Xie et al. · pku

Spatial intelligence is essential for multimodal large language models (MLLMs) operating in the complex physical world. Existing benchmarks, however, probe only single-image relations and thus fail to assess the multi-image spatial reasoning that real-world deployments demand. We introduce MMSI-Bench, a VQA benchmark dedicated to multi-image spatial intelligence. Six 3D-vision researchers spent more than 300 hours meticulously crafting 1,000 challenging, unambiguous multiple-choice questions from over 120,000 images, each paired with carefully designed distractors and a step-by-step reasoning process. We conduct extensive experiments and thoroughly evaluate 34 open-source and proprietary MLLMs, observing a wide gap: the strongest open-source model attains roughly 30% accuracy and OpenAI's o3 reasoning model reaches 40%, while humans score 97%. These results underscore the challenging nature of MMSI-Bench and the substantial headroom for future research. Leveraging the annotated reasoning processes, we also provide an automated error analysis pipeline that diagnoses four dominant failure modes, including (1) grounding errors, (2) overlap-matching and scene-reconstruction errors, (3) situation-transformation reasoning errors, and (4) spatial-logic errors, offering valuable insights for advancing multi-image spatial intelligence. Project page: https://runsenxu.com/projects/MMSI_Bench .

Kaixiong Gong, Kaituo Feng, Bohao Li et al.

Recently, multimodal large language models (MLLMs), such as GPT-4o, Gemini 1.5 Pro, and Reka Core, have expanded their capabilities to include vision and audio modalities. While these models demonstrate impressive performance across a wide range of audio-visual applications, our proposed DeafTest reveals that MLLMs often struggle with simple tasks humans find trivial: 1) determining which of two sounds is louder, and 2) determining which of two sounds has a higher pitch. Motivated by these observations, we introduce AV-Odyssey Bench, a comprehensive audio-visual benchmark designed to assess whether those MLLMs can truly understand the audio-visual information. This benchmark encompasses 4,555 carefully crafted problems, each incorporating text, visual, and audio components. To successfully infer answers, models must effectively leverage clues from both visual and audio inputs. To ensure precise and objective evaluation of MLLM responses, we have structured the questions as multiple-choice, eliminating the need for human evaluation or LLM-assisted assessment. We benchmark a series of closed-source and open-source models and summarize the observations. By revealing the limitations of current models, we aim to provide useful insight for future dataset collection and model development.

Zhendong Liu, Yuanbi Nie, Yingshui Tan et al.

Benefiting from the powerful capabilities of Large Language Models (LLMs), pre-trained visual encoder models connected to an LLMs can realize Vision Language Models (VLMs). However, existing research shows that the visual modality of VLMs is vulnerable, with attackers easily bypassing LLMs' safety alignment through visual modality features to launch attacks. To address this issue, we enhance the existing VLMs' visual modality safety alignment by adding safety modules, including a safety projector, safety tokens, and a safety head, through a two-stage training process, effectively improving the model's defense against risky images. For example, building upon the LLaVA-v1.5 model, we achieve a safety score of 8.26, surpassing the GPT-4V on the Red Teaming Visual Language Models (RTVLM) benchmark. Our method boasts ease of use, high flexibility, and strong controllability, and it enhances safety while having minimal impact on the model's general performance. Moreover, our alignment strategy also uncovers some possible risky content within commonly used open-source multimodal datasets. Our code will be open sourced after the anonymous review.

Yilei Jiang, Xinyan Gao, Tianshuo Peng et al.

The integration of additional modalities increases the susceptibility of large vision-language models (LVLMs) to safety risks, such as jailbreak attacks, compared to their language-only counterparts. While existing research primarily focuses on post-hoc alignment techniques, the underlying safety mechanisms within LVLMs remain largely unexplored. In this work , we investigate whether LVLMs inherently encode safety-relevant signals within their internal activations during inference. Our findings reveal that LVLMs exhibit distinct activation patterns when processing unsafe prompts, which can be leveraged to detect and mitigate adversarial inputs without requiring extensive fine-tuning. Building on this insight, we introduce HiddenDetect, a novel tuning-free framework that harnesses internal model activations to enhance safety. Experimental results show that {HiddenDetect} surpasses state-of-the-art methods in detecting jailbreak attacks against LVLMs. By utilizing intrinsic safety-aware patterns, our method provides an efficient and scalable solution for strengthening LVLM robustness against multimodal threats. Our code will be released publicly at https://github.com/leigest519/HiddenDetect.

Kaixuan Fan, Kaituo Feng, Haoming Lyu et al.

Recent advances have shown success in eliciting strong reasoning abilities in multimodal large language models (MLLMs) through rule-based reinforcement learning (RL) with outcome rewards. However, this paradigm typically lacks supervision over the thinking process leading to the final outcome.As a result, the model may learn sub-optimal reasoning strategies, which can hinder its generalization ability. In light of this, we propose SophiaVL-R1, as an attempt to add reward signals for the thinking process in this paradigm. To achieve this, we first train a thinking reward model that evaluates the quality of the entire thinking process. Given that the thinking reward may be unreliable for certain samples due to reward hacking, we propose the Trust-GRPO method, which assigns a trustworthiness weight to the thinking reward during training. This weight is computed based on the thinking reward comparison of responses leading to correct answers versus incorrect answers, helping to mitigate the impact of potentially unreliable thinking rewards. Moreover, we design an annealing training strategy that gradually reduces the thinking reward over time, allowing the model to rely more on the accurate rule-based outcome reward in later training stages. Experiments show that our SophiaVL-R1 surpasses a series of reasoning MLLMs on various benchmarks (e.g., MathVisita, MMMU), demonstrating strong reasoning and generalization capabilities. Notably, our SophiaVL-R1-7B even outperforms LLaVA-OneVision-72B on most benchmarks, despite the latter having 10 times more parameters. All code, models, and datasets are made publicly available at https://github.com/kxfan2002/SophiaVL-R1.

Xuehui Wang, Zhenyu Wu, JingJing Xie et al. · pku

We introduce MMBench-GUI, a hierarchical benchmark for evaluating GUI automation agents across Windows, macOS, Linux, iOS, Android, and Web platforms. It comprises four levels: GUI Content Understanding, Element Grounding, Task Automation, and Task Collaboration, covering essential skills for GUI agents. In addition, we propose a novel Efficiency-Quality Area (EQA) metric to assess GUI agent execution efficiency in online automation scenarios. Through MMBench-GUI, we identify accurate visual grounding as a critical determinant of overall task success, emphasizing the substantial benefits of modular frameworks that integrate specialized grounding modules. Furthermore, to achieve reliable GUI automation, an agent requires strong task planning and cross-platform generalization abilities, with long-context memory, a broad action space, and long-term reasoning playing a critical role. More important, task efficiency remains a critically underexplored dimension, and all models suffer from substantial inefficiencies, with excessive redundant steps even when tasks are ultimately completed. The integration of precise localization, effective planning, and early stopping strategies is indispensable to enable truly efficient and scalable GUI automation. Our benchmark code, evaluation data, and running environment will be publicly available at https://github.com/open-compass/MMBench-GUI.

Zeqiang Lai, Yunfei Zhao, Zibo Zhao et al.

3D shape generation has greatly flourished through the development of so-called "native" 3D diffusion, particularly through the Vecset Diffusion Model (VDM). While recent advancements have shown promising results in generating high-resolution 3D shapes, VDM still struggles with high-speed generation. Challenges exist because of difficulties not only in accelerating diffusion sampling but also VAE decoding in VDM, areas under-explored in previous works. To address these challenges, we present FlashVDM, a systematic framework for accelerating both VAE and DiT in VDM. For DiT, FlashVDM enables flexible diffusion sampling with as few as 5 inference steps and comparable quality, which is made possible by stabilizing consistency distillation with our newly introduced Progressive Flow Distillation. For VAE, we introduce a lightning vecset decoder equipped with Adaptive KV Selection, Hierarchical Volume Decoding, and Efficient Network Design. By exploiting the locality of the vecset and the sparsity of shape surface in the volume, our decoder drastically lowers FLOPs, minimizing the overall decoding overhead. We apply FlashVDM to Hunyuan3D-2 to obtain Hunyuan3D-2 Turbo. Through systematic evaluation, we show that our model significantly outperforms existing fast 3D generation methods, achieving comparable performance to the state-of-the-art while reducing inference time by over 45x for reconstruction and 32x for generation. Code and models are available at https://github.com/Tencent/FlashVDM.

Kaituo Feng, Manyuan Zhang, Hongyu Li et al.

Reinforcement learning (RL) has recently achieved remarkable success in eliciting visual reasoning within Multimodal Large Language Models (MLLMs). However, existing approaches typically train separate models for different tasks and treat image and video reasoning as disjoint domains. This results in limited scalability toward a multimodal reasoning generalist, which restricts practical versatility and hinders potential knowledge sharing across tasks and modalities. To this end, we propose OneThinker, an all-in-one reasoning model that unifies image and video understanding across diverse fundamental visual tasks, including question answering, captioning, spatial and temporal grounding, tracking, and segmentation. To achieve this, we construct the OneThinker-600k training corpus covering all these tasks and employ commercial models for CoT annotation, resulting in OneThinker-SFT-340k for SFT cold start. Furthermore, we propose EMA-GRPO to handle reward heterogeneity in multi-task RL by tracking task-wise moving averages of reward standard deviations for balanced optimization. Extensive experiments on diverse visual benchmarks show that OneThinker delivers strong performance on 31 benchmarks, across 10 fundamental visual understanding tasks. Moreover, it exhibits effective knowledge transfer between certain tasks and preliminary zero-shot generalization ability, marking a step toward a unified multimodal reasoning generalist. All code, model, and data are released.

Jinmei Liu, Wenbin Li, Xiangyu Yue et al.

We study continual offline reinforcement learning, a practical paradigm that facilitates forward transfer and mitigates catastrophic forgetting to tackle sequential offline tasks. We propose a dual generative replay framework that retains previous knowledge by concurrent replay of generated pseudo-data. First, we decouple the continual learning policy into a diffusion-based generative behavior model and a multi-head action evaluation model, allowing the policy to inherit distributional expressivity for encompassing a progressive range of diverse behaviors. Second, we train a task-conditioned diffusion model to mimic state distributions of past tasks. Generated states are paired with corresponding responses from the behavior generator to represent old tasks with high-fidelity replayed samples. Finally, by interleaving pseudo samples with real ones of the new task, we continually update the state and behavior generators to model progressively diverse behaviors, and regularize the multi-head critic via behavior cloning to mitigate forgetting. Experiments demonstrate that our method achieves better forward transfer with less forgetting, and closely approximates the results of using previous ground-truth data due to its high-fidelity replay of the sample space. Our code is available at \href{https://github.com/NJU-RL/CuGRO}{https://github.com/NJU-RL/CuGRO}.

Yiyuan Zhang, Yuhao Kang, Zhixin Zhang et al.

We introduce $\textit{InteractiveVideo}$, a user-centric framework for video generation. Different from traditional generative approaches that operate based on user-provided images or text, our framework is designed for dynamic interaction, allowing users to instruct the generative model through various intuitive mechanisms during the whole generation process, e.g. text and image prompts, painting, drag-and-drop, etc. We propose a Synergistic Multimodal Instruction mechanism, designed to seamlessly integrate users' multimodal instructions into generative models, thus facilitating a cooperative and responsive interaction between user inputs and the generative process. This approach enables iterative and fine-grained refinement of the generation result through precise and effective user instructions. With $\textit{InteractiveVideo}$, users are given the flexibility to meticulously tailor key aspects of a video. They can paint the reference image, edit semantics, and adjust video motions until their requirements are fully met. Code, models, and demo are available at https://github.com/invictus717/InteractiveVideo

Zhaoyang Liu, Jingjing Xie, Zichen Ding et al.

Vision-Language Models (VLMs) have enabled computer use agents (CUAs) that operate GUIs autonomously, showing great potential, yet progress is limited by the lack of large-scale, open-source computer use data and foundation models. In this work, we introduce ScaleCUA, a step toward scaling open-source CUAs. It offers a large-scale dataset spanning 6 operating systems and 3 task domains, built via a closed-loop pipeline uniting automated agents with human experts. Trained on this scaled-up data, ScaleCUA can operate seamlessly across platforms. Specifically, it delivers strong gains over baselines (+26.6 on WebArena-Lite-v2, +10.7 on ScreenSpot-Pro) and sets new state-of-the-art results (94.4% on MMBench-GUI L1-Hard, 60.6% on OSWorld-G, 47.4% on WebArena-Lite-v2). These findings underscore the power of data-driven scaling for general-purpose computer use agents. We will release data, models, and code to advance future research: https://github.com/OpenGVLab/ScaleCUA.

Yaxuan Huang, Xili Dai, Jianan Wang et al.

Room layout estimation from multiple-perspective images is poorly investigated due to the complexities that emerge from multi-view geometry, which requires muti-step solutions such as camera intrinsic and extrinsic estimation, image matching, and triangulation. However, in 3D reconstruction, the advancement of recent 3D foundation models such as DUSt3R has shifted the paradigm from the traditional multi-step structure-from-motion process to an end-to-end single-step approach. To this end, we introduce Plane-DUSt3R, a novel method for multi-view room layout estimation leveraging the 3D foundation model DUSt3R. Plane-DUSt3R incorporates the DUSt3R framework and fine-tunes on a room layout dataset (Structure3D) with a modified objective to estimate structural planes. By generating uniform and parsimonious results, Plane-DUSt3R enables room layout estimation with only a single post-processing step and 2D detection results. Unlike previous methods that rely on single-perspective or panorama image, Plane-DUSt3R extends the setting to handle multiple-perspective images. Moreover, it offers a streamlined, end-to-end solution that simplifies the process and reduces error accumulation. Experimental results demonstrate that Plane-DUSt3R not only outperforms state-of-the-art methods on the synthetic dataset but also proves robust and effective on in the wild data with different image styles such as cartoon. Our code is available at: https://github.com/justacar/Plane-DUSt3R

Yinan Xia, Yilei Jiang, Yingshui Tan et al.

Vision-Language Models (VLMs) have achieved remarkable progress in multimodal reasoning tasks through enhanced chain-of-thought capabilities. However, this advancement also introduces novel safety risks, as these models become increasingly vulnerable to harmful multimodal prompts that can trigger unethical or unsafe behaviors. Existing safety alignment approaches, primarily designed for unimodal language models, fall short in addressing the complex and nuanced threats posed by multimodal inputs. Moreover, current safety datasets lack the fine-grained, policy-grounded reasoning required to robustly align reasoning-capable VLMs. In this work, we introduce {MSR-Align}, a high-quality Multimodal Safety Reasoning dataset tailored to bridge this gap. MSR-Align supports fine-grained, deliberative reasoning over standardized safety policies across both vision and text modalities. Our data generation pipeline emphasizes multimodal diversity, policy-grounded reasoning, and rigorous quality filtering using strong multimodal judges. Extensive experiments demonstrate that fine-tuning VLMs on MSR-Align substantially improves robustness against both textual and vision-language jailbreak attacks, while preserving or enhancing general reasoning performance. MSR-Align provides a scalable and effective foundation for advancing the safety alignment of reasoning-capable VLMs. Our dataset is made publicly available at https://huggingface.co/datasets/Leigest/MSR-Align.

Chuang Yu, Jinmiao Zhao, Mingxuan Zhao et al.

Recently, multimodal large language models (MLLMs) have been widely applied to reasoning tasks. However, they suffer from limited multi-rationale semantic modeling, insufficient logical robustness, and are susceptible to misleading interpretations in complex scenarios. Therefore, we propose a Multi-rationale INtegrated Discriminative (MIND) reasoning framework, which is designed to endow MLLMs with human-like cognitive abilities of "Understand -> Rethink -> Correct", and achieves a paradigm evolution from passive imitation-based reasoning to active discriminative reasoning. Specifically, we introduce a Rationale Augmentation and Discrimination (RAD) paradigm, which automatically and efficiently expands existing datasets by generating diverse rationales, providing a unified and extensible data foundation. Meanwhile, we design a Progressive Two-stage Correction Learning (P2CL) strategy. The first phase enhances multi-rationale positive learning, while the second phase enables active logic discrimination and correction. In addition, to mitigate representation entanglement in the multi-rationale semantic space, we propose a Multi-rationale Contrastive Alignment (MCA) optimization strategy, which achieves semantic aggregation of correct reasoning and boundary separation of incorrect reasoning. Extensive experiments demonstrate that the proposed MIND reasoning framework achieves state-of-the-art (SOTA) performance on multiple public datasets covering scientific, commonsense, and mathematical scenarios. It provides a new perspective for advancing MLLMs towards higher levels of cognitive intelligence. Our code is available at https://github.com/YuChuang1205/MIND

Chuang Yu, Jinmiao Zhao, Yunpeng Liu et al.

While large-scale visual foundation models (VFMs) exhibit strong generalization across diverse visual domains, their potential for single-frame infrared small target (SIRST) detection remains largely unexplored. To fill this gap, we systematically introduce the frozen representations from VFMs into the SIRST task for the first time and propose a Foundation-Driven Efficient Paradigm (FDEP), which can seamlessly adapt to existing encoder-decoder-based methods and significantly improve accuracy without additional inference overhead. Specifically, a Semantic Alignment Modulation Fusion (SAMF) module is designed to achieve dynamic alignment and deep fusion of the global semantic priors from VFMs with task-specific features. Meanwhile, to avoid the inference time burden introduced by VFMs, we propose a Collaborative Optimization-based Implicit Self-Distillation (CO-ISD) strategy, which enables implicit semantic transfer between the main and lightweight branches through parameter sharing and synchronized backpropagation. In addition, to unify the fragmented evaluation system, we construct a Holistic SIRST Evaluation (HSE) metric that performs multi-threshold integral evaluation at both pixel-level confidence and target-level robustness, providing a stable and comprehensive basis for fair model comparison. Extensive experiments demonstrate that the SIRST detection networks equipped with our FDEP framework achieve state-of-the-art (SOTA) performance on multiple public datasets. Our code is available at https://github.com/YuChuang1205/FDEP-Framework

Qunzhong Wang, Jie Liu, Jiajun Liang et al.

Recent advancements in multimodal reward models (RMs) have substantially improved post-training for visual generative models. However, current RMs face inherent limitations: (1) visual inputs consume large context budgets, forcing fewer frames and causing loss of fine-grained details; and (2) all visual information is packed into the initial prompt, exacerbating hallucination and forgetting during chain-of-thought reasoning. To overcome these issues, we introduce VideoReward Thinker (VR-Thinker), a thinking-with-image framework that equips the RM with visual reasoning operations (e.g., select frame) and a configurable visual memory window. This allows the RM to actively acquire and update visual evidence within context limits, improving reasoning fidelity and reliability. We activate visual reasoning via a reinforcement fine-tuning pipeline: (i) Cold Start with curated visual chain-of-thought data to distill basic reasoning skills and operation formatting; (ii) select samples whose per-dimension and overall judgments are all correct, then conduct Rejection sampling Fine-Tuning on these high-quality traces to further enhance reasoning; and (iii) apply Group Relative Policy Optimization (GRPO) to strengthen reasoning. Our approach delivers state-of-the-art accuracy among open-source models on video preference benchmarks, especially for longer videos: a 7B VR-Thinker achieves 80.5% on VideoGen Reward, 82.3% on GenAI-Bench, and 75.6% on MJ-Bench-Video. These results validate the effectiveness and promise of thinking-with-image multimodal reward modeling.

Haoran Hao, Jiaming Han, Yiyuan Zhang et al. · cmu

Recent advances in Large Language Models (LLMs) have led to significant breakthroughs in video understanding. However, existing models still struggle with long video processing due to the context length constraint of LLMs and the vast amount of information within the video. Although some recent methods are designed for long video understanding, they often lose crucial information during token compression and struggle with additional modality like audio. In this work, we propose a dynamic long video encoding method utilizing the temporal relationship between frames, named Temporal Dynamic Context (TDC). Firstly, we segment the video into semantically consistent scenes based on inter-frame similarities, then encode each frame into tokens using visual-audio encoders. Secondly, we propose a novel temporal context compressor to reduce the number of tokens within each segment. Specifically, we employ a query-based Transformer to aggregate video, audio, and instruction text tokens into a limited set of temporal context tokens. Finally, we feed the static frame tokens and the temporal context tokens into the LLM for video understanding. Furthermore, to handle extremely long videos, we propose a training-free chain-of-thought strategy that progressively extracts answers from multiple video segments. These intermediate answers serve as part of the reasoning process and contribute to the final answer. We conduct extensive experiments on general video understanding and audio-video understanding benchmarks, where our method demonstrates strong performance. The code and models are available at https://github.com/Hoar012/TDC-Video.

Yiyuan Zhang, Handong Li, Jing Liu et al.

We propose to build omni-modal intelligence, which is capable of understanding any modality and learning universal representations. In specific, we propose a scalable pretraining paradigm, named Multimodal Context (MiCo), which can scale up the numbers of modalities and amount of data, together with the model parameters, in the pretraining process. With MiCo, the pretrained models show significant emergent abilities in multimodal learning, which are evaluated on the following tasks: i) single-modality perception benchmarks of 10 different modalities, ii) 25 cross-modality understanding tasks of retrieval, question-answering, captioning, and iii) 18 multimodal large language model benchmarks. Our models establish 37 new records for state-of-the-art performance. We hope that our research could contribute to the development of omni-modal intelligence. Code and Models are at https://github.com/invictus717/MiCo

Chen Tang, Xinzhu Ma, Encheng Su et al.

Traditional spatiotemporal models generally rely on task-specific architectures, which limit their generalizability and scalability across diverse tasks due to domain-specific design requirements. In this paper, we introduce \textbf{UniSTD}, a unified Transformer-based framework for spatiotemporal modeling, which is inspired by advances in recent foundation models with the two-stage pretraining-then-adaption paradigm. Specifically, our work demonstrates that task-agnostic pretraining on 2D vision and vision-text datasets can build a generalizable model foundation for spatiotemporal learning, followed by specialized joint training on spatiotemporal datasets to enhance task-specific adaptability. To improve the learning capabilities across domains, our framework employs a rank-adaptive mixture-of-expert adaptation by using fractional interpolation to relax the discrete variables so that can be optimized in the continuous space. Additionally, we introduce a temporal module to incorporate temporal dynamics explicitly. We evaluate our approach on a large-scale dataset covering 10 tasks across 4 disciplines, demonstrating that a unified spatiotemporal model can achieve scalable, cross-task learning and support up to 10 tasks simultaneously within one model while reducing training costs in multi-domain applications. Code will be available at https://github.com/1hunters/UniSTD.

Chuang Yu, Yunpeng Liu, Jinmiao Zhao et al.

Recently, cross-spectral image patch matching based on feature relation learning has attracted extensive attention. However, performance bottleneck problems have gradually emerged in existing methods. To address this challenge, we make the first attempt to explore a stable and efficient bridge between descriptor learning and metric learning, and construct a knowledge-guided learning network (KGL-Net), which achieves amazing performance improvements while abandoning complex network structures. Specifically, we find that there is feature extraction consistency between metric learning based on feature difference learning and descriptor learning based on Euclidean distance. This provides the foundation for bridge building. To ensure the stability and efficiency of the constructed bridge, on the one hand, we conduct an in-depth exploration of 20 combined network architectures. On the other hand, a feature-guided loss is constructed to achieve mutual guidance of features. In addition, unlike existing methods, we consider that the feature mapping ability of the metric branch should receive more attention. Therefore, a hard negative sample mining for metric learning (HNSM-M) strategy is constructed. To the best of our knowledge, this is the first time that hard negative sample mining for metric networks has been implemented and brings significant performance gains. Extensive experimental results show that our KGL-Net achieves SOTA performance in three different cross-spectral image patch matching scenarios. Our code are available at https://github.com/YuChuang1205/KGL-Net.

Yiyuan Zhang, Xiaohan Ding, Kaixiong Gong et al.

We propose to improve transformers of a specific modality with irrelevant data from other modalities, e.g., improve an ImageNet model with audio or point cloud datasets. We would like to highlight that the data samples of the target modality are irrelevant to the other modalities, which distinguishes our method from other works utilizing paired (e.g., CLIP) or interleaved data of different modalities. We propose a methodology named Multimodal Pathway - given a target modality and a transformer designed for it, we use an auxiliary transformer trained with data of another modality and construct pathways to connect components of the two models so that data of the target modality can be processed by both models. In this way, we utilize the universal sequence-to-sequence modeling abilities of transformers obtained from two modalities. As a concrete implementation, we use a modality-specific tokenizer and task-specific head as usual but utilize the transformer blocks of the auxiliary model via a proposed method named Cross-Modal Re-parameterization, which exploits the auxiliary weights without any inference costs. On the image, point cloud, video, and audio recognition tasks, we observe significant and consistent performance improvements with irrelevant data from other modalities. The code and models are available at https://github.com/AILab-CVC/M2PT.

Chenfeng Xu, Shijia Yang, Tomer Galanti et al.

3D point-clouds and 2D images are different visual representations of the physical world. While human vision can understand both representations, computer vision models designed for 2D image and 3D point-cloud understanding are quite different. Our paper explores the potential of transferring 2D model architectures and weights to understand 3D point-clouds, by empirically investigating the feasibility of the transfer, the benefits of the transfer, and shedding light on why the transfer works. We discover that we can indeed use the same architecture and pretrained weights of a neural net model to understand both images and point-clouds. Specifically, we transfer the image-pretrained model to a point-cloud model by copying or inflating the weights. We find that finetuning the transformed image-pretrained models (FIP) with minimal efforts -- only on input, output, and normalization layers -- can achieve competitive performance on 3D point-cloud classification, beating a wide range of point-cloud models that adopt task-specific architectures and use a variety of tricks. When finetuning the whole model, the performance improves even further. Meanwhile, FIP improves data efficiency, reaching up to 10.0 top-1 accuracy percent on few-shot classification. It also speeds up the training of point-cloud models by up to 11.1x for a target accuracy (e.g., 90 % accuracy). Lastly, we provide an explanation of the image to point-cloud transfer from the aspect of neural collapse. The code is available at: \url{https://github.com/chenfengxu714/image2point}.

Sicheng Zhao, Yang Xiao, Jiang Guo et al.

Sentiment analysis of user-generated reviews or comments on products and services in social networks can help enterprises to analyze the feedback from customers and take corresponding actions for improvement. To mitigate large-scale annotations on the target domain, domain adaptation (DA) provides an alternate solution by learning a transferable model from other labeled source domains. Existing multi-source domain adaptation (MDA) methods either fail to extract some discriminative features in the target domain that are related to sentiment, neglect the correlations of different sources and the distribution difference among different sub-domains even in the same source, or cannot reflect the varying optimal weighting during different training stages. In this paper, we propose a novel instance-level MDA framework, named curriculum cycle-consistent generative adversarial network (C-CycleGAN), to address the above issues. Specifically, C-CycleGAN consists of three components: (1) pre-trained text encoder which encodes textual input from different domains into a continuous representation space, (2) intermediate domain generator with curriculum instance-level adaptation which bridges the gap across source and target domains, and (3) task classifier trained on the intermediate domain for final sentiment classification. C-CycleGAN transfers source samples at instance-level to an intermediate domain that is closer to the target domain with sentiment semantics preserved and without losing discriminative features. Further, our dynamic instance-level weighting mechanisms can assign the optimal weights to different source samples in each training stage. We conduct extensive experiments on three benchmark datasets and achieve substantial gains over state-of-the-art DA approaches. Our source code is released at: https://github.com/WArushrush/Curriculum-CycleGAN.

Hanchen Wang, Qi Liu, Xiangyu Yue et al.

We describe a simple pre-training approach for point clouds. It works in three steps: 1. Mask all points occluded in a camera view; 2. Learn an encoder-decoder model to reconstruct the occluded points; 3. Use the encoder weights as initialisation for downstream point cloud tasks. We find that even when we construct a single pre-training dataset (from ModelNet40), this pre-training method improves accuracy across different datasets and encoders, on a wide range of downstream tasks. Specifically, we show that our method outperforms previous pre-training methods in object classification, and both part-based and semantic segmentation tasks. We study the pre-trained features and find that they lead to wide downstream minima, have high transformation invariance, and have activations that are highly correlated with part labels. Code and data are available at: https://github.com/hansen7/OcCo

Sicheng Zhao, Bo Li, Xiangyu Yue et al.

Simulation-to-real domain adaptation for semantic segmentation has been actively studied for various applications such as autonomous driving. Existing methods mainly focus on a single-source setting, which cannot easily handle a more practical scenario of multiple sources with different distributions. In this paper, we propose to investigate multi-source domain adaptation for semantic segmentation. Specifically, we design a novel framework, termed Multi-source Adversarial Domain Aggregation Network (MADAN), which can be trained in an end-to-end manner. First, we generate an adapted domain for each source with dynamic semantic consistency while aligning at the pixel-level cycle-consistently towards the target. Second, we propose sub-domain aggregation discriminator and cross-domain cycle discriminator to make different adapted domains more closely aggregated. Finally, feature-level alignment is performed between the aggregated domain and target domain while training the segmentation network. Extensive experiments from synthetic GTA and SYNTHIA to real Cityscapes and BDDS datasets demonstrate that the proposed MADAN model outperforms state-of-the-art approaches. Our source code is released at: https://github.com/Luodian/MADAN.

Bichen Wu, Xuanyu Zhou, Sicheng Zhao et al.

Earlier work demonstrates the promise of deep-learning-based approaches for point cloud segmentation; however, these approaches need to be improved to be practically useful. To this end, we introduce a new model SqueezeSegV2 that is more robust to dropout noise in LiDAR point clouds. With improved model structure, training loss, batch normalization and additional input channel, SqueezeSegV2 achieves significant accuracy improvement when trained on real data. Training models for point cloud segmentation requires large amounts of labeled point-cloud data, which is expensive to obtain. To sidestep the cost of collection and annotation, simulators such as GTA-V can be used to create unlimited amounts of labeled, synthetic data. However, due to domain shift, models trained on synthetic data often do not generalize well to the real world. We address this problem with a domain-adaptation training pipeline consisting of three major components: 1) learned intensity rendering, 2) geodesic correlation alignment, and 3) progressive domain calibration. When trained on real data, our new model exhibits segmentation accuracy improvements of 6.0-8.6% over the original SqueezeSeg. When training our new model on synthetic data using the proposed domain adaptation pipeline, we nearly double test accuracy on real-world data, from 29.0% to 57.4%. Our source code and synthetic dataset will be open-sourced.

Bichen Wu, Alvin Wan, Xiangyu Yue et al.

In this paper, we address semantic segmentation of road-objects from 3D LiDAR point clouds. In particular, we wish to detect and categorize instances of interest, such as cars, pedestrians and cyclists. We formulate this problem as a point- wise classification problem, and propose an end-to-end pipeline called SqueezeSeg based on convolutional neural networks (CNN): the CNN takes a transformed LiDAR point cloud as input and directly outputs a point-wise label map, which is then refined by a conditional random field (CRF) implemented as a recurrent layer. Instance-level labels are then obtained by conventional clustering algorithms. Our CNN model is trained on LiDAR point clouds from the KITTI dataset, and our point-wise segmentation labels are derived from 3D bounding boxes from KITTI. To obtain extra training data, we built a LiDAR simulator into Grand Theft Auto V (GTA-V), a popular video game, to synthesize large amounts of realistic training data. Our experiments show that SqueezeSeg achieves high accuracy with astonishingly fast and stable runtime (8.7 ms per frame), highly desirable for autonomous driving applications. Furthermore, additionally training on synthesized data boosts validation accuracy on real-world data. Our source code and synthesized data will be open-sourced.

Di Zhang, Wei Liu, Qian Tan et al.

Large language models (LLMs) have made impressive progress in chemistry applications. However, the community lacks an LLM specifically designed for chemistry. The main challenges are two-fold: firstly, most chemical data and scientific knowledge are stored in structured databases, which limits the model's ability to sustain coherent dialogue when used directly. Secondly, there is an absence of objective and fair benchmark that encompass most chemistry tasks. Here, we introduce ChemLLM, a comprehensive framework that features the first LLM dedicated to chemistry. It also includes ChemData, a dataset specifically designed for instruction tuning, and ChemBench, a robust benchmark covering nine essential chemistry tasks. ChemLLM is adept at performing various tasks across chemical disciplines with fluid dialogue interaction. Notably, ChemLLM achieves results comparable to GPT-4 on the core chemical tasks and demonstrates competitive performance with LLMs of similar size in general scenarios. ChemLLM paves a new path for exploration in chemical studies, and our method of incorporating structured chemical knowledge into dialogue systems sets a new standard for developing LLMs in various scientific fields. Codes, Datasets, and Model weights are publicly accessible at https://hf.co/AI4Chem

Chenyu Huang, Peng Ye, Tao Chen et al.

The success of pretrain-finetune paradigm brings about the release of numerous model weights. In this case, merging models finetuned on different tasks to enable a single model with multi-task capabilities is gaining increasing attention for its practicability. Existing model merging methods usually suffer from (1) significant performance degradation or (2) requiring tuning by additional data or training. In this paper, we rethink and analyze the existing model merging paradigm. We discover that using a single model's weights can hardly simulate all the models' performance. To tackle this issue, we propose Elect, Mask & Rescale-Merging (EMR-Merging). We first (a) elect a unified model from all the model weights and then (b) generate extremely lightweight task-specific modulators, including masks and rescalers, to align the direction and magnitude between the unified model and each specific model, respectively. EMR-Merging is tuning-free, thus requiring no data availability or any additional training while showing impressive performance. We find that EMR-Merging shows outstanding performance compared to existing merging methods under different classical and newly-established settings, including merging different numbers of vision models (up to 30), NLP models, PEFT models, and multi-modal models.