Zilong Huang

Wenqiang Zhang, Zilong Huang, Guozhong Luo et al. · deepmind, tencent-ai

Although vision transformers (ViTs) have achieved great success in computer vision, the heavy computational cost hampers their applications to dense prediction tasks such as semantic segmentation on mobile devices. In this paper, we present a mobile-friendly architecture named \textbf{To}ken \textbf{P}yramid Vision Trans\textbf{former} (\textbf{TopFormer}). The proposed \textbf{TopFormer} takes Tokens from various scales as input to produce scale-aware semantic features, which are then injected into the corresponding tokens to augment the representation. Experimental results demonstrate that our method significantly outperforms CNN- and ViT-based networks across several semantic segmentation datasets and achieves a good trade-off between accuracy and latency. On the ADE20K dataset, TopFormer achieves 5\% higher accuracy in mIoU than MobileNetV3 with lower latency on an ARM-based mobile device. Furthermore, the tiny version of TopFormer achieves real-time inference on an ARM-based mobile device with competitive results. The code and models are available at: https://github.com/hustvl/TopFormer

Qiang Wan, Zilong Huang, Jiachen Lu et al. · tencent-ai

Since the introduction of Vision Transformers, the landscape of many computer vision tasks (e.g., semantic segmentation), which has been overwhelmingly dominated by CNNs, recently has significantly revolutionized. However, the computational cost and memory requirement renders these methods unsuitable on the mobile device. In this paper, we introduce a new method squeeze-enhanced Axial Transformer (SeaFormer) for mobile visual recognition. Specifically, we design a generic attention block characterized by the formulation of squeeze Axial and detail enhancement. It can be further used to create a family of backbone architectures with superior cost-effectiveness. Coupled with a light segmentation head, we achieve the best trade-off between segmentation accuracy and latency on the ARM-based mobile devices on the ADE20K, Cityscapes, Pascal Context and COCO-Stuff datasets. Critically, we beat both the mobilefriendly rivals and Transformer-based counterparts with better performance and lower latency without bells and whistles. Furthermore, we incorporate a feature upsampling-based multi-resolution distillation technique, further reducing the inference latency of the proposed framework. Beyond semantic segmentation, we further apply the proposed SeaFormer architecture to image classification and object detection problems, demonstrating the potential of serving as a versatile mobile-friendly backbone. Our code and models are made publicly available at https://github.com/fudan-zvg/SeaFormer.

Xin Chen, Biao Jiang, Wen Liu et al. · deepmind, tencent-ai

We study a challenging task, conditional human motion generation, which produces plausible human motion sequences according to various conditional inputs, such as action classes or textual descriptors. Since human motions are highly diverse and have a property of quite different distribution from conditional modalities, such as textual descriptors in natural languages, it is hard to learn a probabilistic mapping from the desired conditional modality to the human motion sequences. Besides, the raw motion data from the motion capture system might be redundant in sequences and contain noises; directly modeling the joint distribution over the raw motion sequences and conditional modalities would need a heavy computational overhead and might result in artifacts introduced by the captured noises. To learn a better representation of the various human motion sequences, we first design a powerful Variational AutoEncoder (VAE) and arrive at a representative and low-dimensional latent code for a human motion sequence. Then, instead of using a diffusion model to establish the connections between the raw motion sequences and the conditional inputs, we perform a diffusion process on the motion latent space. Our proposed Motion Latent-based Diffusion model (MLD) could produce vivid motion sequences conforming to the given conditional inputs and substantially reduce the computational overhead in both the training and inference stages. Extensive experiments on various human motion generation tasks demonstrate that our MLD achieves significant improvements over the state-of-the-art methods among extensive human motion generation tasks, with two orders of magnitude faster than previous diffusion models on raw motion sequences.

Wenkai Wang, Tao Xiong, Jingchen Ni et al.

Real-world professional desktop workflows in specialized creative and engineering software unfold over long horizons and often require human-in-the-loop coordination, where agents proactively seek necessary information and users provide additional instructions, clarifications, feedback, or corrections as the task progresses. Yet existing desktop GUI benchmarks mostly reduce this setting to short, simplified tasks with all user instructions provided upfront. To address this issue, we introduce DeskCraft, a desktop GUI benchmark targeting long horizon creative and engineering workflows and proactive human-agent collaboration. DeskCraft organizes tasks into a multilevel difficulty taxonomy, with long horizon tasks requiring over 50 execution steps, and covers professional creative software across design, video, audio, and 3D creation. Furthermore, DeskCraft formalizes human-agent collaboration into an interaction protocol covering mid-turn and post-turn exchanges. Mid-turn interaction captures both agent-initiated clarification under uncertainty and user-initiated interruption during execution, while post-turn interaction accommodates user-driven feedback after the agent signals completion, together spanning the full space of realistic collaboration patterns. We evaluate 18 proprietary and open source agents on 538 tasks and find that GPT-5.4 reaches 31.6% on standard tasks and 27.6% on interactive tasks. Further analyses reveal persistent failures in long horizon workflow delivery and proactive clarification. We will open-source all evaluation codes, tasks, and data at https://github.com/mrwwk/DeskCraft.

Fukun Yin, Wen Liu, Zilong Huang et al. · deepmind, tencent-ai

Implicit neural 3D representation has achieved impressive results in surface or scene reconstruction and novel view synthesis, which typically uses the coordinate-based multi-layer perceptrons (MLPs) to learn a continuous scene representation. However, existing approaches, such as Neural Radiance Field (NeRF) and its variants, usually require dense input views (i.e. 50-150) to obtain decent results. To relive the over-dependence on massive calibrated images and enrich the coordinate-based feature representation, we explore injecting the prior information into the coordinate-based network and introduce a novel coordinate-based model, CoCo-INR, for implicit neural 3D representation. The cores of our method are two attention modules: codebook attention and coordinate attention. The former extracts the useful prototypes containing rich geometry and appearance information from the prior codebook, and the latter propagates such prior information into each coordinate and enriches its feature representation for a scene or object surface. With the help of the prior information, our method can render 3D views with more photo-realistic appearance and geometries than the current methods using fewer calibrated images available. Experiments on various scene reconstruction datasets, including DTU and BlendedMVS, and the full 3D head reconstruction dataset, H3DS, demonstrate the robustness under fewer input views and fine detail-preserving capability of our proposed method.

Andrey Ignatov, Grigory Malivenko, Radu Timofte et al. · tencent-ai

Various depth estimation models are now widely used on many mobile and IoT devices for image segmentation, bokeh effect rendering, object tracking and many other mobile tasks. Thus, it is very crucial to have efficient and accurate depth estimation models that can run fast on low-power mobile chipsets. In this Mobile AI challenge, the target was to develop deep learning-based single image depth estimation solutions that can show a real-time performance on IoT platforms and smartphones. For this, the participants used a large-scale RGB-to-depth dataset that was collected with the ZED stereo camera capable to generated depth maps for objects located at up to 50 meters. The runtime of all models was evaluated on the Raspberry Pi 4 platform, where the developed solutions were able to generate VGA resolution depth maps at up to 27 FPS while achieving high fidelity results. All models developed in the challenge are also compatible with any Android or Linux-based mobile devices, their detailed description is provided in this paper.

Yanda Li, Zilong Huang, Gang Yu et al. · tencent-ai

Image matting requires high-quality pixel-level human annotations to support the training of a deep model in recent literature. Whereas such annotation is costly and hard to scale, significantly holding back the development of the research. In this work, we make the first attempt towards addressing this problem, by proposing a self-supervised pre-training approach that can leverage infinite numbers of data to boost the matting performance. The pre-training task is designed in a similar manner as image matting, where random trimap and alpha matte are generated to achieve an image disentanglement objective. The pre-trained model is then used as an initialisation of the downstream matting task for fine-tuning. Extensive experimental evaluations show that the proposed approach outperforms both the state-of-the-art matting methods and other alternative self-supervised initialisation approaches by a large margin. We also show the robustness of the proposed approach over different backbone architectures. Our project page is available at https://crystraldo.github.io/dpt_mat/.

Yang Zhao, Zhijie Lin, Daquan Zhou et al.

LLMs have demonstrated remarkable abilities at interacting with humans through language, especially with the usage of instruction-following data. Recent advancements in LLMs, such as MiniGPT-4, LLaVA, and X-LLM, further enlarge their abilities by incorporating multi-modal inputs, including image, video, and speech. Despite their effectiveness at generating precise and detailed language understanding of the given modality signal, these LLMs give up the ability to ground specific parts of inputs, thus only constructing a coarse-grained mapping. However, explicit and informative correspondence between text and other modalities will not only improve the user experience but also help to expand the application scenario of multi-modal LLMs. Therefore, we propose BuboGPT, a multi-modal LLM with visual grounding that can perform cross-modal interaction between vision, audio and language, providing fine-grained understanding of visual objects and other given modalities. As a result, BuboGPT is able to point out the specific location of an object in the image, when it is generating response or description for that object. Our contributions are two-fold: 1) An off-the-shelf visual grounding module based on SAM that extracts entities in a sentence and find corresponding masks in the image. 2) A two-stage training scheme and instruction dataset to endow joint text-image-audio understanding. Our experiments show that BuboGPT achieves impressive multi-modality understanding and visual grounding abilities during the interaction with human. It performs consistently well when provided by arbitrary modality combinations (either aligned or unaligned). Our code, model and dataset are available at https://bubo-gpt.github.io .

Lianghui Zhu, Yuxin Fang, Bencheng Liao et al.

Scaling depth is a key driver for large language models (LLMs). Yet, as LLMs become deeper, they often suffer from signal degradation: informative features formed in shallow layers are gradually diluted by repeated residual updates, making them harder to recover in deeper layers. We introduce mixture-of-depths attention (MoDA), a mechanism that allows each attention head to attend to sequence KV pairs at the current layer and depth KV pairs from preceding layers. We further describe a hardware-efficient algorithm for MoDA that resolves non-contiguous memory-access patterns, achieving 97.3% of FlashAttention-2's efficiency at a sequence length of 64K. Experiments on 1.5B-parameter models demonstrate that MoDA consistently outperforms strong baselines. Notably, it improves average perplexity by 0.2 across 10 validation benchmarks and increases average performance by 2.11% on 10 downstream tasks, with a negligible 3.7% FLOPs computational overhead. We also find that combining MoDA with post-norm yields better performance than using it with pre-norm. These results suggest that MoDA is a promising primitive for depth scaling. Code is released at https://github.com/hustvl/MoDA .

Qirui Shen, Wenda Wang, Jiachen Lu et al.

Architectural spatial intelligence, the ability to recognize and infer architectural space, is fundamental to tasks such as robot navigation, embodied interaction, and 3D scene understanding and generation. Although extensive research has evaluated the basic spatial skills of Vision-Language Models (VLMs) such as relative orientation, distance comparison, and object counting, these tasks cover only the most elementary levels of spatial cognition and largely overlook higher-level cognition of architectural space, including layout understanding, circulation patterns, and functional zoning. In this work, we present ArchSIBench, a Benchmark for Architectural Spatial Intelligence based on the perspectives from architecture, cognitive science, and psychology. ArchSIBench covers five core dimensions: perception, reasoning, navigation, transformation, and configuration, comprising 17 fine-grained subtasks. Through careful manual annotation by experts with architectural backgrounds, we construct 3,000 question-answer pairs to enable comprehensive evaluation of architectural spatial intelligence. Based on ArchSIBench, we evaluate various VLMs and find that the architectural spatial intelligence of most models shows significant differences from human baselines; additionally, models exhibit substantial variability across capability dimensions. Some state-of-the-art models can approach the level of human evaluators without architectural training. However, a clear gap remains compared to human evaluators with architectural training, particularly in spatial transformation and configuration reasoning. We believe that ArchSIBench will provide important insights and systematic resources for measuring and advancing the architectural spatial intelligence of VLMs. The dataset and code are available at https://huggingface.co/datasets/ArchSIBench/ArchSIBench.

Weiheng Zhao, Zilong Huang, Jiashi Feng et al.

Contrastive Language-Image Pretraining (CLIP) achieves strong generalization in vision-language tasks by aligning images and texts in a shared embedding space. However, recent findings show that CLIP-like models still underutilize fine-grained semantic signals in text, and this issue becomes even more pronounced when dealing with long and detailed captions. This stems from CLIP's training objective, which optimizes only global image-text similarity and overlooks token-level supervision - limiting its ability to achieve fine-grained visual-text alignment. To address this, we propose SuperCLIP, a simple yet effective framework that augments contrastive learning with classification-based supervision. By adding only a lightweight linear layer to the vision encoder, SuperCLIP leverages token-level cues to enhance visual-textual alignment - with just a 0.077% increase in total FLOPs, and no need for additional annotated data. Experiments show that SuperCLIP consistently improves zero-shot classification, image-text retrieval, and purely visual tasks. These gains hold regardless of whether the model is trained on original web data or rich re-captioned data, demonstrating SuperCLIP's ability to recover textual supervision in both cases. Furthermore, SuperCLIP alleviates CLIP's small-batch performance drop through classification-based supervision that avoids reliance on large batch sizes. Code and models will be made open source.

Junyan Ye, Jun He, Zilong Huang et al.

Image generation models have evolved from text-conditioned pixel synthesis toward multimodal agents endowed with visual comprehension and tool invocation capabilities. Yet, existing agents remain at the mercy of underlying black-box image models. Their workflow is trapped in a repetitive cycle of prompt rewriting for generation refinement, leaving them with no mechanism to directly manipulate the canvas. In essence, the potential of LLMs to serve as a genuine "brush" for precise visual construction remains largely untapped. In this paper, we propose GenClaw, a code-driven agentic image generation paradigm that empowers the agent to create like a human artist: first conceptualizing, then sketching, and finally coloring. Specifically, the agent first constructs the conceptual knowledge and context through search and reasoning. It then utilizes code (e.g., SVG, HTML, Three.js) to render executable visual sketches. Finally, it employs an image generation model to supplement textures, materials, and photorealism. In this workflow, code serves as a controllable intermediate canvas bridging linguistic reasoning and pixel synthesis, seamlessly integrating programmatic logic with the visual expressiveness of generative models. By transforming image generation from a black-box paradigm into a staged process akin to authentic human creation, GenClaw offers a step toward for highly controllable and interpretable visual generation systems.

Qi Song, Honglin Li, Yingchen Yu et al.

Recent releases such as o3 highlight human-like "thinking with images" reasoning that combines structured tool use with stepwise verification, yet most open-source approaches still rely on text-only chains, rigid visual schemas, or single-step pipelines, limiting flexibility, interpretability, and transferability on complex tasks. We introduce CodeDance, which explores executable code as a general solver for visual reasoning. Unlike fixed-schema calls (e.g., only predicting bounding-box coordinates), CodeDance defines, composes, and executes code to orchestrate multiple tools, compute intermediate results, and render visual artifacts (e.g., boxes, lines, plots) that support transparent, self-checkable reasoning. To guide this process, we introduce a reward for balanced and adaptive tool-call, which balances exploration with efficiency and mitigates tool overuse. Interestingly, beyond the expected capabilities taught by atomic supervision, we empirically observe novel emergent behaviors during RL training: CodeDance demonstrates novel tool invocations, unseen compositions, and cross-task transfer. These behaviors arise without task-specific fine-tuning, suggesting a general and scalable mechanism of executable visual reasoning. Extensive experiments across reasoning benchmarks (e.g., visual search, math, chart QA) show that CodeDance not only consistently outperforms schema-driven and text-only baselines, but also surpasses advanced closed models such as GPT-4o and larger open-source models.

Weijia Li, Jun He, Junyan Ye et al.

Satellite-to-street view synthesis aims at generating a realistic street-view image from its corresponding satellite-view image. Although stable diffusion models have exhibit remarkable performance in a variety of image generation applications, their reliance on similar-view inputs to control the generated structure or texture restricts their application to the challenging cross-view synthesis task. In this work, we propose CrossViewDiff, a cross-view diffusion model for satellite-to-street view synthesis. To address the challenges posed by the large discrepancy across views, we design the satellite scene structure estimation and cross-view texture mapping modules to construct the structural and textural controls for street-view image synthesis. We further design a cross-view control guided denoising process that incorporates the above controls via an enhanced cross-view attention module. To achieve a more comprehensive evaluation of the synthesis results, we additionally design a GPT-based scoring method as a supplement to standard evaluation metrics. We also explore the effect of different data sources (e.g., text, maps, building heights, and multi-temporal satellite imagery) on this task. Results on three public cross-view datasets show that CrossViewDiff outperforms current state-of-the-art on both standard and GPT-based evaluation metrics, generating high-quality street-view panoramas with more realistic structures and textures across rural, suburban, and urban scenes. The code and models of this work will be released at https://opendatalab.github.io/CrossViewDiff/.

Siyu Jiao, Yiheng Lin, Yujie Zhong et al.

Recent progress in Multimodal Large Language Models (MLLMs) demonstrates that Chain-of-Thought (CoT) reasoning enables systematic solutions to complex understanding tasks. However, its extension to generation tasks remains nascent and limited by scenario-specific mechanisms that hinder generalization and adaptation. In this work, we present ThinkGen, the first think-driven visual generation framework that explicitly leverages MLLM's CoT reasoning in various generation scenarios. ThinkGen employs a decoupled architecture comprising a pretrained MLLM and a Diffusion Transformer (DiT), wherein the MLLM generates tailored instructions based on user intent, and DiT produces high-quality images guided by these instructions. We further propose a separable GRPO-based training paradigm (SepGRPO), alternating reinforcement learning between the MLLM and DiT modules. This flexible design enables joint training across diverse datasets, facilitating effective CoT reasoning for a wide range of generative scenarios. Extensive experiments demonstrate that ThinkGen achieves robust, state-of-the-art performance across multiple generation benchmarks. Code is available: https://github.com/jiaosiyuu/ThinkGen

Haobo Yuan, Xiangtai Li, Tao Zhang et al.

This work presents Sa2VA, the first comprehensive, unified model for dense grounded understanding of both images and videos. Unlike existing multi-modal large language models, which are often limited to specific modalities and tasks, Sa2VA supports a wide range of image and video tasks, including referring segmentation and conversation, with minimal one-shot instruction tuning. Sa2VA combines SAM-2, a foundation video segmentation model, with MLLM, the advanced vision-language model, and unifies text, image, and video into a shared LLM token space. Using the LLM, Sa2VA generates instruction tokens that guide SAM-2 in producing precise masks, enabling a grounded, multi-modal understanding of both static and dynamic visual content. Additionally, we introduce Ref-SAV, an auto-labeled dataset containing over 72k object expressions in complex video scenes, designed to boost model performance. We also manually validate 2k video objects in the Ref-SAV datasets to benchmark referring video object segmentation in complex environments. Experiments show that Sa2VA achieves strong performance across multiple tasks, particularly in referring video object segmentation, highlighting its potential for complex real-world applications. In addition, Sa2VA can be easily extended into various VLMs, including Qwen-VL and Intern-VL, which can be updated with rapid process in current open-sourced VLMs. Code and models have been provided to the community.

Tianwei Xiong, Jun Hao Liew, Zilong Huang et al.

Autoregressive (AR) video generative models rely on video tokenizers that compress pixels into discrete token sequences. The length of these token sequences is crucial for balancing reconstruction quality against downstream generation computational cost. Traditional video tokenizers apply a uniform token assignment across temporal blocks of different videos, often wasting tokens on simple, static, or repetitive segments while underserving dynamic or complex ones. To address this inefficiency, we introduce $\textbf{EVATok}$, a framework to produce $\textbf{E}$fficient $\textbf{V}$ideo $\textbf{A}$daptive $\textbf{Tok}$enizers. Our framework estimates optimal token assignments for each video to achieve the best quality-cost trade-off, develops lightweight routers for fast prediction of these optimal assignments, and trains adaptive tokenizers that encode videos based on the assignments predicted by routers. We demonstrate that EVATok delivers substantial improvements in efficiency and overall quality for video reconstruction and downstream AR generation. Enhanced by our advanced training recipe that integrates video semantic encoders, EVATok achieves superior reconstruction and state-of-the-art class-to-video generation on UCF-101, with at least 24.4% savings in average token usage compared to the prior state-of-the-art LARP and our fixed-length baseline.

Zhiyuan Yan, Junyan Ye, Weijia Li et al.

The recent breakthroughs in OpenAI's GPT4o model have demonstrated surprisingly good capabilities in image generation and editing, resulting in significant excitement in the community. This technical report presents the first-look evaluation benchmark (named GPT-ImgEval), quantitatively and qualitatively diagnosing GPT-4o's performance across three critical dimensions: (1) generation quality, (2) editing proficiency, and (3) world knowledge-informed semantic synthesis. Across all three tasks, GPT-4o demonstrates strong performance, significantly surpassing existing methods in both image generation control and output quality, while also showcasing exceptional knowledge reasoning capabilities. Furthermore, based on the GPT-4o's generated data, we propose a classification-model-based approach to investigate the underlying architecture of GPT-4o, where our empirical results suggest the model consists of an auto-regressive (AR) combined with a diffusion-based head for image decoding, rather than the VAR-like architectures. We also provide a complete speculation on GPT-4o's overall architecture. In addition, we conduct a series of analyses to identify and visualize GPT-4o's specific limitations and the synthetic artifacts commonly observed in its image generation. We also present a comparative study of multi-round image editing between GPT-4o and Gemini 2.0 Flash, and discuss the safety implications of GPT-4o's outputs, particularly their detectability by existing image forensic models. We hope that our work can offer valuable insight and provide a reliable benchmark to guide future research, foster reproducibility, and accelerate innovation in the field of image generation and beyond. The codes and datasets used for evaluating GPT-4o can be found at https://github.com/PicoTrex/GPT-ImgEval.

Junyan Ye, Baichuan Zhou, Zilong Huang et al.

With the rapid development of AI-generated content, the future internet may be inundated with synthetic data, making the discrimination of authentic and credible multimodal data increasingly challenging. Synthetic data detection has thus garnered widespread attention, and the performance of large multimodal models (LMMs) in this task has attracted significant interest. LMMs can provide natural language explanations for their authenticity judgments, enhancing the explainability of synthetic content detection. Simultaneously, the task of distinguishing between real and synthetic data effectively tests the perception, knowledge, and reasoning capabilities of LMMs. In response, we introduce LOKI, a novel benchmark designed to evaluate the ability of LMMs to detect synthetic data across multiple modalities. LOKI encompasses video, image, 3D, text, and audio modalities, comprising 18K carefully curated questions across 26 subcategories with clear difficulty levels. The benchmark includes coarse-grained judgment and multiple-choice questions, as well as fine-grained anomaly selection and explanation tasks, allowing for a comprehensive analysis of LMMs. We evaluated 22 open-source LMMs and 6 closed-source models on LOKI, highlighting their potential as synthetic data detectors and also revealing some limitations in the development of LMM capabilities. More information about LOKI can be found at https://opendatalab.github.io/LOKI/

Tao Zhang, Xiangtai Li, Zilong Huang et al.

Multimodal Large Language Models (MLLMs) achieve remarkable performance for fine-grained pixel-level understanding tasks. However, all the works rely heavily on extra components, such as vision encoder (CLIP), segmentation experts, leading to high system complexity and limiting model scaling. In this work, our goal is to explore a highly simplified MLLM without introducing extra components. Our work is motivated by the recent works on Single trAnsformer as a unified vIsion-Language Model (SAIL) design, where these works jointly learn vision tokens and text tokens in transformers. We present Pixel-SAIL, a single transformer for pixel-wise MLLM tasks. In particular, we present three technical improvements on the plain baseline. First, we design a learnable upsampling module to refine visual token features. Secondly, we propose a novel visual prompt injection strategy to enable the single transformer to understand visual prompt inputs and benefit from the early fusion of visual prompt embeddings and vision tokens. Thirdly, we introduce a vision expert distillation strategy to efficiently enhance the single transformer's fine-grained feature extraction capability. In addition, we have collected a comprehensive pixel understanding benchmark (PerBench), using a manual check. It includes three tasks: detailed object description, visual prompt-based question answering, and visual-text referring segmentation. Extensive experiments on four referring segmentation benchmarks, one visual prompt benchmark, and our PerBench show that our Pixel-SAIL achieves comparable or even better results with a much simpler pipeline. Code and model will be released at https://github.com/magic-research/Sa2VA.

Weixian Lei, Jiacong Wang, Haochen Wang et al.

This paper introduces SAIL, a single transformer unified multimodal large language model (MLLM) that integrates raw pixel encoding and language decoding within a singular architecture. Unlike existing modular MLLMs, which rely on a pre-trained vision transformer (ViT), SAIL eliminates the need for a separate vision encoder, presenting a more minimalist architecture design. Instead of introducing novel architectural components, SAIL adapts mix-attention mechanisms and multimodal positional encodings to better align with the distinct characteristics of visual and textual modalities. We systematically compare SAIL's properties-including scalability, cross-modal information flow patterns, and visual representation capabilities-with those of modular MLLMs. By scaling both training data and model size, SAIL achieves performance comparable to modular MLLMs. Notably, the removal of pretrained ViT components enhances SAIL's scalability and results in significantly different cross-modal information flow patterns. Moreover, SAIL demonstrates strong visual representation capabilities, achieving results on par with ViT-22B in vision tasks such as semantic segmentation. Code and models are available at https://github.com/bytedance/SAIL.

Junyan Ye, Dongzhi Jiang, Zihao Wang et al.

Recently, GPT-4o has garnered significant attention for its strong performance in image generation, yet open-source models still lag behind. Several studies have explored distilling image data from GPT-4o to enhance open-source models, achieving notable progress. However, a key question remains: given that real-world image datasets already constitute a natural source of high-quality data, why should we use GPT-4o-generated synthetic data? In this work, we identify two key advantages of synthetic images. First, they can complement rare scenarios in real-world datasets, such as surreal fantasy or multi-reference image generation, which frequently occur in user queries. Second, they provide clean and controllable supervision. Real-world data often contains complex background noise and inherent misalignment between text descriptions and image content, whereas synthetic images offer pure backgrounds and long-tailed supervision signals, facilitating more accurate text-to-image alignment. Building on these insights, we introduce Echo-4o-Image, a 180K-scale synthetic dataset generated by GPT-4o, harnessing the power of synthetic image data to address blind spots in real-world coverage. Using this dataset, we fine-tune the unified multimodal generation baseline Bagel to obtain Echo-4o. In addition, we propose two new evaluation benchmarks for a more accurate and challenging assessment of image generation capabilities: GenEval++, which increases instruction complexity to mitigate score saturation, and Imagine-Bench, which focuses on evaluating both the understanding and generation of imaginative content. Echo-4o demonstrates strong performance across standard benchmarks. Moreover, applying Echo-4o-Image to other foundation models (e.g., OmniGen2, BLIP3-o) yields consistent performance gains across multiple metrics, highlighting the datasets strong transferability.

Jun He, Junyan Ye, Zilong Huang et al.

While text-to-image generation has achieved unprecedented fidelity, the vast majority of existing models function fundamentally as static text-to-pixel decoders. Consequently, they often fail to grasp implicit user intentions. Although emerging unified understanding-generation models have improved intent comprehension, they still struggle to accomplish tasks involving complex knowledge reasoning within a single model. Moreover, constrained by static internal priors, these models remain unable to adapt to the evolving dynamics of the real world. To bridge these gaps, we introduce Mind-Brush, a unified agentic framework that transforms generation into a dynamic, knowledge-driven workflow. Simulating a human-like 'think-research-create' paradigm, Mind-Brush actively retrieves multimodal evidence to ground out-of-distribution concepts and employs reasoning tools to resolve implicit visual constraints. To rigorously evaluate these capabilities, we propose Mind-Bench, a comprehensive benchmark comprising 500 distinct samples spanning real-time news, emerging concepts, and domains such as mathematical and Geo-Reasoning. Extensive experiments demonstrate that Mind-Brush significantly enhances the capabilities of unified models, realizing a zero-to-one capability leap for the Qwen-Image baseline on Mind-Bench, while achieving superior results on established benchmarks like WISE and RISE.

Haochen Wang, Xiangtai Li, Zilong Huang et al.

Models like OpenAI-o3 pioneer visual grounded reasoning by dynamically referencing visual regions, just like human "thinking with images". However, no benchmark exists to evaluate these capabilities holistically. To bridge this gap, we propose TreeBench (Traceable Evidence Evaluation Benchmark), a diagnostic benchmark built on three principles: (1) focused visual perception of subtle targets in complex scenes, (2) traceable evidence via bounding box evaluation, and (3) second-order reasoning to test object interactions and spatial hierarchies beyond simple object localization. Prioritizing images with dense objects, we initially sample 1K high-quality images from SA-1B, and incorporate eight LMM experts to manually annotate questions, candidate options, and answers for each image. After three stages of quality control, TreeBench consists of 405 challenging visual question-answering pairs, even the most advanced models struggle with this benchmark, where none of them reach 60% accuracy, e.g., OpenAI-o3 scores only 54.87. Furthermore, we introduce TreeVGR (Traceable Evidence Enhanced Visual Grounded Reasoning), a training paradigm to supervise localization and reasoning jointly with reinforcement learning, enabling accurate localizations and explainable reasoning pathways. Initialized from Qwen2.5-VL-7B, it improves V* Bench (+16.8), MME-RealWorld (+12.6), and TreeBench (+13.4), proving traceability is key to advancing vision-grounded reasoning. The code is available at https://github.com/Haochen-Wang409/TreeVGR.

Zilong Huang, Qinghao Ye, Bingyi Kang et al.

We introduce SuperClass, a super simple classification method for vision-language pre-training on image-text data. Unlike its contrastive counterpart CLIP who contrast with a text encoder, SuperClass directly utilizes tokenized raw text as supervised classification labels, without the need for additional text filtering or selection. Due to the absence of the text encoding as contrastive target, SuperClass does not require a text encoder and does not need to maintain a large batch size as CLIP does. SuperClass demonstrated superior performance on various downstream tasks, including classic computer vision benchmarks and vision language downstream tasks. We further explored the scaling behavior of SuperClass on model size, training length, or data size, and reported encouraging results and comparisons to CLIP. https://github.com/x-cls/superclass

Zilong Huang, Jun He, Junyan Ye et al.

The reconstruction of immersive and realistic 3D scenes holds significant practical importance in various fields of computer vision and computer graphics. Typically, immersive and realistic scenes should be free from obstructions by dynamic objects, maintain global texture consistency, and allow for unrestricted exploration. The current mainstream methods for image-driven scene construction involves iteratively refining the initial image using a moving virtual camera to generate the scene. However, previous methods struggle with visual discontinuities due to global texture inconsistencies under varying camera poses, and they frequently exhibit scene voids caused by foreground-background occlusions. To this end, we propose a novel layered 3D scene reconstruction framework from panoramic image, named Scene4U. Specifically, Scene4U integrates an open-vocabulary segmentation model with a large language model to decompose a real panorama into multiple layers. Then, we employs a layered repair module based on diffusion model to restore occluded regions using visual cues and depth information, generating a hierarchical representation of the scene. The multi-layer panorama is then initialized as a 3D Gaussian Splatting representation, followed by layered optimization, which ultimately produces an immersive 3D scene with semantic and structural consistency that supports free exploration. Scene4U outperforms state-of-the-art method, improving by 24.24% in LPIPS and 24.40% in BRISQUE, while also achieving the fastest training speed. Additionally, to demonstrate the robustness of Scene4U and allow users to experience immersive scenes from various landmarks, we build WorldVista3D dataset for 3D scene reconstruction, which contains panoramic images of globally renowned sites. The implementation code and dataset will be released at https://github.com/LongHZ140516/Scene4U .

Dong Guo, Faming Wu, Feida Zhu et al. · pku

We present Seed1.5-VL, a vision-language foundation model designed to advance general-purpose multimodal understanding and reasoning. Seed1.5-VL is composed with a 532M-parameter vision encoder and a Mixture-of-Experts (MoE) LLM of 20B active parameters. Despite its relatively compact architecture, it delivers strong performance across a wide spectrum of public VLM benchmarks and internal evaluation suites, achieving the state-of-the-art performance on 38 out of 60 public benchmarks. Moreover, in agent-centric tasks such as GUI control and gameplay, Seed1.5-VL outperforms leading multimodal systems, including OpenAI CUA and Claude 3.7. Beyond visual and video understanding, it also demonstrates strong reasoning abilities, making it particularly effective for multimodal reasoning challenges such as visual puzzles. We believe these capabilities will empower broader applications across diverse tasks. In this report, we mainly provide a comprehensive review of our experiences in building Seed1.5-VL across model design, data construction, and training at various stages, hoping that this report can inspire further research. Seed1.5-VL is now accessible at https://www.volcengine.com/ (Volcano Engine Model ID: doubao-1-5-thinking-vision-pro-250428)

Junyan Ye, Dongzhi Jiang, Jun He et al.

Recently, Multimodal Large Language Models (MLLMs) have made rapid progress, particularly in enhancing their reasoning capabilities. However, existing reasoning benchmarks still primarily assess language-based reasoning, often treating visual input as replaceable context. To address this gap, we introduce BLINK-Twice, a vision-centric reasoning benchmark grounded in challenging perceptual tasks. Instead of relying on external knowledge, our tasks require models to reason from visual content alone, shifting the focus from language-based to image-grounded reasoning. Compared to prior perception benchmarks, it moves beyond shallow perception ("see") and requires fine-grained observation and analytical reasoning ("observe"). BLINK-Twice integrates three core components: seven types of visual challenges for testing visual reasoning, natural adversarial image pairs that enforce reliance on visual content, and annotated reasoning chains for fine-grained evaluation of the reasoning process rather than final answers alone. We evaluate 20 leading MLLMs, including 12 foundation models and 8 reasoning-enhanced models. BLINK-Twice poses a significant challenge to current models. While existing reasoning strategies in the language space-such as chain-of-thought or self-criticism can improve performance, they often result in unstable and redundant reasoning. We observe that repeated image observation improves performance across models, and active visual interaction, as demonstrated by models like o3, highlights the need for a new paradigm for vision reasoning. The dataset is publicly available at https://github.com/PicoTrex/BLINK-Twice

Mingsheng Li, Lin Zhang, Mingzhen Zhu et al.

This paper studies the problem of pre-training for small models, which is essential for many mobile devices. Current state-of-the-art methods on this problem transfer the representational knowledge of a large network (as a Teacher) into a smaller model (as a Student) using self-supervised distillation, improving the performance of the small model on downstream tasks. However, existing approaches are insufficient in extracting the crucial knowledge that is useful for discerning categories in downstream tasks during the distillation process. In this paper, for the first time, we introduce language guidance to the distillation process and propose a new method named Language-Guided Distillation (LGD) system, which uses category names of the target downstream task to help refine the knowledge transferred between the teacher and student. To this end, we utilize a pre-trained text encoder to extract semantic embeddings from language and construct a textual semantic space called Textual Semantics Bank (TSB). Furthermore, we design a Language-Guided Knowledge Aggregation (LGKA) module to construct the visual semantic space, also named Visual Semantics Bank (VSB). The task-related knowledge is transferred by driving a student encoder to mimic the similarity score distribution inferred by a teacher over TSB and VSB. Compared with other small models obtained by either ImageNet pre-training or self-supervised distillation, experiment results show that the distilled lightweight model using the proposed LGD method presents state-of-the-art performance and is validated on various downstream tasks, including classification, detection, and segmentation. We have made the code available at https://github.com/mZhenz/LGD.

Lihe Yang, Bingyi Kang, Zilong Huang et al.

This work presents Depth Anything, a highly practical solution for robust monocular depth estimation. Without pursuing novel technical modules, we aim to build a simple yet powerful foundation model dealing with any images under any circumstances. To this end, we scale up the dataset by designing a data engine to collect and automatically annotate large-scale unlabeled data (~62M), which significantly enlarges the data coverage and thus is able to reduce the generalization error. We investigate two simple yet effective strategies that make data scaling-up promising. First, a more challenging optimization target is created by leveraging data augmentation tools. It compels the model to actively seek extra visual knowledge and acquire robust representations. Second, an auxiliary supervision is developed to enforce the model to inherit rich semantic priors from pre-trained encoders. We evaluate its zero-shot capabilities extensively, including six public datasets and randomly captured photos. It demonstrates impressive generalization ability. Further, through fine-tuning it with metric depth information from NYUv2 and KITTI, new SOTAs are set. Our better depth model also results in a better depth-conditioned ControlNet. Our models are released at https://github.com/LiheYoung/Depth-Anything.

Jitesh Jain, Anukriti Singh, Nikita Orlov et al.

Finetuning a pretrained backbone in the encoder part of an image transformer network has been the traditional approach for the semantic segmentation task. However, such an approach leaves out the semantic context that an image provides during the encoding stage. This paper argues that incorporating semantic information of the image into pretrained hierarchical transformer-based backbones while finetuning improves the performance considerably. To achieve this, we propose SeMask, a simple and effective framework that incorporates semantic information into the encoder with the help of a semantic attention operation. In addition, we use a lightweight semantic decoder during training to provide supervision to the intermediate semantic prior maps at every stage. Our experiments demonstrate that incorporating semantic priors enhances the performance of the established hierarchical encoders with a slight increase in the number of FLOPs. We provide empirical proof by integrating SeMask into Swin Transformer and Mix Transformer backbones as our encoder paired with different decoders. Our framework achieves a new state-of-the-art of 58.25% mIoU on the ADE20K dataset and improvements of over 3% in the mIoU metric on the Cityscapes dataset. The code and checkpoints are publicly available at https://github.com/Picsart-AI-Research/SeMask-Segmentation .

Qiang Zhou, Zilong Huang, Lichao Huang et al.

Video object segmentation (VOS) aims at pixel-level object tracking given only the annotations in the first frame. Due to the large visual variations of objects in video and the lack of training samples, it remains a difficult task despite the upsurging development of deep learning. Toward solving the VOS problem, we bring in several new insights by the proposed unified framework consisting of object proposal, tracking and segmentation components. The object proposal network transfers objectness information as generic knowledge into VOS; the tracking network identifies the target object from the proposals; and the segmentation network is performed based on the tracking results with a novel dynamic-reference based model adaptation scheme. Extensive experiments have been conducted on the DAVIS'17 dataset and the YouTube-VOS dataset, our method achieves the state-of-the-art performance on several video object segmentation benchmarks. We make the code publicly available at https://github.com/sydney0zq/PTSNet.

Zilong Huang, Xinggang Wang, Yunchao Wei et al.

Contextual information is vital in visual understanding problems, such as semantic segmentation and object detection. We propose a Criss-Cross Network (CCNet) for obtaining full-image contextual information in a very effective and efficient way. Concretely, for each pixel, a novel criss-cross attention module harvests the contextual information of all the pixels on its criss-cross path. By taking a further recurrent operation, each pixel can finally capture the full-image dependencies. Besides, a category consistent loss is proposed to enforce the criss-cross attention module to produce more discriminative features. Overall, CCNet is with the following merits: 1) GPU memory friendly. Compared with the non-local block, the proposed recurrent criss-cross attention module requires 11x less GPU memory usage. 2) High computational efficiency. The recurrent criss-cross attention significantly reduces FLOPs by about 85% of the non-local block. 3) The state-of-the-art performance. We conduct extensive experiments on semantic segmentation benchmarks including Cityscapes, ADE20K, human parsing benchmark LIP, instance segmentation benchmark COCO, video segmentation benchmark CamVid. In particular, our CCNet achieves the mIoU scores of 81.9%, 45.76% and 55.47% on the Cityscapes test set, the ADE20K validation set and the LIP validation set respectively, which are the new state-of-the-art results. The source codes are available at \url{https://github.com/speedinghzl/CCNet}.

Tao Ruan, Ting Liu, Zilong Huang et al.

Human parsing has received considerable interest due to its wide application potentials. Nevertheless, it is still unclear how to develop an accurate human parsing system in an efficient and elegant way. In this paper, we identify several useful properties, including feature resolution, global context information and edge details, and perform rigorous analyses to reveal how to leverage them to benefit the human parsing task. The advantages of these useful properties finally result in a simple yet effective Context Embedding with Edge Perceiving (CE2P) framework for single human parsing. Our CE2P is end-to-end trainable and can be easily adopted for conducting multiple human parsing. Benefiting the superiority of CE2P, we achieved the 1st places on all three human parsing benchmarks. Without any bells and whistles, we achieved 56.50\% (mIoU), 45.31\% (mean $AP^r$) and 33.34\% ($AP^p_{0.5}$) in LIP, CIHP and MHP v2.0, which outperform the state-of-the-arts more than 2.06\%, 3.81\% and 1.87\%, respectively. We hope our CE2P will serve as a solid baseline and help ease future research in single/multiple human parsing. Code has been made available at \url{https://github.com/liutinglt/CE2P}.

Sili Chen, Hengkai Guo, Shengnan Zhu et al.

Depth Anything has achieved remarkable success in monocular depth estimation with strong generalization ability. However, it suffers from temporal inconsistency in videos, hindering its practical applications. Various methods have been proposed to alleviate this issue by leveraging video generation models or introducing priors from optical flow and camera poses. Nonetheless, these methods are only applicable to short videos (< 10 seconds) and require a trade-off between quality and computational efficiency. We propose Video Depth Anything for high-quality, consistent depth estimation in super-long videos (over several minutes) without sacrificing efficiency. We base our model on Depth Anything V2 and replace its head with an efficient spatial-temporal head. We design a straightforward yet effective temporal consistency loss by constraining the temporal depth gradient, eliminating the need for additional geometric priors. The model is trained on a joint dataset of video depth and unlabeled images, similar to Depth Anything V2. Moreover, a novel key-frame-based strategy is developed for long video inference. Experiments show that our model can be applied to arbitrarily long videos without compromising quality, consistency, or generalization ability. Comprehensive evaluations on multiple video benchmarks demonstrate that our approach sets a new state-of-the-art in zero-shot video depth estimation. We offer models of different scales to support a range of scenarios, with our smallest model capable of real-time performance at 30 FPS.

Yan Fang, Mengcheng Lan, Zilong Huang et al.

In this paper, we present \textbf{Gen}erative \textbf{L}anguage-\textbf{I}mage \textbf{P}re-training (GenLIP), a minimalist generative pretraining framework for Vision Transformers (ViTs) designed for multimodal large language models (MLLMs). To better align vision encoders with the autoregressive nature of LLMs, GenLIP trains a ViT to predict language tokens directly from visual tokens using a standard language modeling objective, without contrastive batch construction or an additional text decoder. This design offers three key advantages: (1) \textbf{Simplicity}: a single transformer jointly models visual and textual tokens; (2) \textbf{Scalability}: it scales effectively with both data and model size; and (3) \textbf{Performance}: it achieves competitive or superior results across diverse multimodal benchmarks. Trained on 8B samples from Recap-DataComp-1B, GenLIP matches or surpasses strong baselines despite using substantially less pretraining data. After continued pretraining on multi-resolution images at native aspect ratios, GenLIP further improves on detail-sensitive tasks such as OCR and chart understanding, making it a strong foundation for vision encoders in MLLMs.

Qiang Wan, Zilong Huang, Bingyi Kang et al.

The issue of generative pretraining for vision models has persisted as a long-standing conundrum. At present, the text-to-image (T2I) diffusion model demonstrates remarkable proficiency in generating high-definition images matching textual inputs, a feat made possible through its pre-training on large-scale image-text pairs. This leads to a natural inquiry: can diffusion models be utilized to tackle visual perception tasks? In this paper, we propose a simple yet effective scheme to harness a diffusion model for visual perception tasks. Our key insight is to introduce learnable embeddings (meta prompts) to the pre-trained diffusion models to extract proper features for perception. The effect of meta prompts are two-fold. First, as a direct replacement of the text embeddings in the T2I models, it can activate task-relevant features during feature extraction. Second, it will be used to re-arrange the extracted features to ensures that the model focuses on the most pertinent features for the task on hand. Additionally, we design a recurrent refinement training strategy that fully leverages the property of diffusion models, thereby yielding stronger visual features. Extensive experiments across various benchmarks validate the effectiveness of our approach. Our approach achieves new performance records in depth estimation tasks on NYU depth V2 and KITTI, and in semantic segmentation task on CityScapes. Concurrently, the proposed method attains results comparable to the current state-of-the-art in semantic segmentation on ADE20K and pose estimation on COCO datasets, further exemplifying its robustness and versatility.

Tianwei Xiong, Jun Hao Liew, Zilong Huang et al.

In autoregressive (AR) image generation, visual tokenizers compress images into compact discrete latent tokens, enabling efficient training of downstream autoregressive models for visual generation via next-token prediction. While scaling visual tokenizers improves image reconstruction quality, it often degrades downstream generation quality -- a challenge not adequately addressed in existing literature. To address this, we introduce GigaTok, the first approach to simultaneously improve image reconstruction, generation, and representation learning when scaling visual tokenizers. We identify the growing complexity of latent space as the key factor behind the reconstruction vs. generation dilemma. To mitigate this, we propose semantic regularization, which aligns tokenizer features with semantically consistent features from a pre-trained visual encoder. This constraint prevents excessive latent space complexity during scaling, yielding consistent improvements in both reconstruction and downstream autoregressive generation. Building on semantic regularization, we explore three key practices for scaling tokenizers:(1) using 1D tokenizers for better scalability, (2) prioritizing decoder scaling when expanding both encoder and decoder, and (3) employing entropy loss to stabilize training for billion-scale tokenizers. By scaling to $\bf{3 \space billion}$ parameters, GigaTok achieves state-of-the-art performance in reconstruction, downstream AR generation, and downstream AR representation quality.

Xiangtai Li, Tao Zhang, Yanwei Li et al.

Multimodal Large Language Models (MLLMs) demonstrate a complex understanding of scenes, benefiting from large-scale and high-quality datasets. Most existing caption datasets lack the ground locations and relations for visual entities. Several grounded caption datasets face the problems of missing detailed descriptions, relations, and massive object descriptions on high-resolution images. To fill this gap for the community, we present DenseWorld-1M, the first massive, detailed, dense grounded caption dataset in the real world. We design a three-stage labeling pipeline, containing open-world perception, detailed object caption generation, and dense caption merging. The first stage obtains entity-level masks and labels. The second stage generates the object-level, detailed captions with the guidance of masks and labels from the first stage. The final stage merges object captions and masks into spatial and relational dense captions. To accelerate the labeling process and improve caption quality, we present two VLM models: the Detailed Region Caption model and the Spatial Caption Merging model. Extensive experiments on various settings, including vision-language understanding, visual grounding, and region caption generation, demonstrate the effectiveness of our DenseWorld-1M dataset and labeling models.

Haobo Yuan, Tao Zhang, Xiangtai Li et al.

Referring video object segmentation (RVOS) is a challenging task that requires the model to segment the object in a video given the language description. MeViS is a recently proposed dataset that contains motion expressions of the target objects, leading to a challenging benchmark, compared with existing RVOS benchmarks. On the other hand, for referring expression tasks, a new trend is to adopt multi-modal large language model (MLLM) to achieve better image and text alignment. In this report, we show that with a simple modification to the test time inference method on stronger MLLMs, we can lead to stronger results on MeVIS. In particular, we adopt the recent method Sa2VA, a unified model for dense grounded understanding of both images and videos. By enlarging the scope of key frames, without any further training, we can achieve the 3rd place in the 4th PVUW workshop.

Jun He, Yi Lin, Zilong Huang et al.

Urban development impacts over half of the global population, making human-centered understanding of its structural and perceptual changes essential for sustainable development. While Multimodal Large Language Models (MLLMs) have shown remarkable capabilities across various domains, existing benchmarks that explore their performance in urban environments remain limited, lacking systematic exploration of temporal evolution and subjective perception of urban environment that aligns with human perception. To address these limitations, we propose UrbanFeel, a comprehensive benchmark designed to evaluate the performance of MLLMs in urban development understanding and subjective environmental perception. UrbanFeel comprises 14.3K carefully constructed visual questions spanning three cognitively progressive dimensions: Static Scene Perception, Temporal Change Understanding, and Subjective Environmental Perception. We collect multi-temporal single-view and panoramic street-view images from 11 representative cities worldwide, and generate high-quality question-answer pairs through a hybrid pipeline of spatial clustering, rule-based generation, model-assisted prompting, and manual annotation. Through extensive evaluation of 20 state-of-the-art MLLMs, we observe that Gemini-2.5 Pro achieves the best overall performance, with its accuracy approaching human expert levels and narrowing the average gap to just 1.5\%. Most models perform well on tasks grounded in scene understanding. In particular, some models even surpass human annotators in pixel-level change detection. However, performance drops notably in tasks requiring temporal reasoning over urban development. Additionally, in the subjective perception dimension, several models reach human-level or even higher consistency in evaluating dimension such as beautiful and safety.

Zilong Huang, Jun He, Xiaobin Huang et al.

Generating realistic 3D cities is fundamental to world models, virtual reality, and game development, where an ideal urban scene must satisfy both stylistic diversity, fine-grained, and controllability. However, existing methods struggle to balance the creative flexibility offered by text-based generation with the object-level editability enabled by explicit structural representations. We introduce MajutsuCity, a natural language-driven and aesthetically adaptive framework for synthesizing structurally consistent and stylistically diverse 3D urban scenes. MajutsuCity represents a city as a composition of controllable layouts, assets, and materials, and operates through a four-stage pipeline. To extend controllability beyond initial generation, we further integrate MajutsuAgent, an interactive language-grounded editing agent} that supports five object-level operations. To support photorealistic and customizable scene synthesis, we also construct MajutsuDataset, a high-quality multimodal dataset} containing 2D semantic layouts and height maps, diverse 3D building assets, and curated PBR materials and skyboxes, each accompanied by detailed annotations. Meanwhile, we develop a practical set of evaluation metrics, covering key dimensions such as structural consistency, scene complexity, material fidelity, and lighting atmosphere. Extensive experiments demonstrate MajutsuCity reduces layout FID by 83.7% compared with CityDreamer and by 20.1% over CityCraft. Our method ranks first across all AQS and RDR scores, outperforming existing methods by a clear margin. These results confirm MajutsuCity as a new state-of-the-art in geometric fidelity, stylistic adaptability, and semantic controllability for 3D city generation. We expect our framework can inspire new avenues of research in 3D city generation. Our project page: https://longhz140516.github.io/MajutsuCity/.

Ruijie Fan, Junyan Ye, Huan Chen et al.

Existing satellite video tracking methods often struggle with generalization, requiring scenario-specific training to achieve satisfactory performance, and are prone to track loss in the presence of occlusion. To address these challenges, we propose SatSAM2, a zero-shot satellite video tracker built on SAM2, designed to adapt foundation models to the remote sensing domain. SatSAM2 introduces two core modules: a Kalman Filter-based Constrained Motion Module (KFCMM) to exploit temporal motion cues and suppress drift, and a Motion-Constrained State Machine (MCSM) to regulate tracking states based on motion dynamics and reliability. To support large-scale evaluation, we propose MatrixCity Video Object Tracking (MVOT), a synthetic benchmark containing 1,500+ sequences and 157K annotated frames with diverse viewpoints, illumination, and occlusion conditions. Extensive experiments on two satellite tracking benchmarks and MVOT show that SatSAM2 outperforms both traditional and foundation model-based trackers, including SAM2 and its variants. Notably, on the OOTB dataset, SatSAM2 achieves a 5.84% AUC improvement over state-of-the-art methods. Our code and dataset will be publicly released to encourage further research.

Haochen Wang, Yuhao Wang, Tao Zhang et al.

While Multimodal Large Language Models (MLLMs) excel at holistic understanding, they struggle in capturing the dense world with complex scenes, requiring fine-grained analysis of intricate details and object inter-relationships. Region-level MLLMs have been a promising step. However, previous attempts are generally optimized to understand given regions in isolation, neglecting crucial global contexts. To address this, we introduce Grasp Any Region (GAR) for comprehen- sive region-level visual understanding. Empowered by an effective RoI-aligned feature replay technique, GAR supports (1) precise perception by leveraging necessary global contexts, and (2) modeling interactions between multiple prompts. Together, it then naturally achieves (3) advanced compositional reasoning to answer specific free-form questions about any region, shifting the paradigm from passive description to active dialogue. Moreover, we construct GAR-Bench, which not only provides a more accurate evaluation of single-region comprehension, but also, more importantly, measures interactions and complex reasoning across multiple regions. Extensive experiments have demonstrated that GAR-1B not only maintains the state-of-the-art captioning capabilities, e.g., outperforming DAM-3B +4.5 on DLC-Bench, but also excels at modeling relationships between multiple prompts with advanced comprehension capabilities, even surpassing InternVL3-78B on GAR-Bench-VQA. More importantly, our zero-shot GAR-8B even outperforms in-domain VideoRefer-7B on VideoRefer-BenchQ, indicating its strong capabilities can be easily transferred to videos.

Lihe Yang, Bingyi Kang, Zilong Huang et al.

This work presents Depth Anything V2. Without pursuing fancy techniques, we aim to reveal crucial findings to pave the way towards building a powerful monocular depth estimation model. Notably, compared with V1, this version produces much finer and more robust depth predictions through three key practices: 1) replacing all labeled real images with synthetic images, 2) scaling up the capacity of our teacher model, and 3) teaching student models via the bridge of large-scale pseudo-labeled real images. Compared with the latest models built on Stable Diffusion, our models are significantly more efficient (more than 10x faster) and more accurate. We offer models of different scales (ranging from 25M to 1.3B params) to support extensive scenarios. Benefiting from their strong generalization capability, we fine-tune them with metric depth labels to obtain our metric depth models. In addition to our models, considering the limited diversity and frequent noise in current test sets, we construct a versatile evaluation benchmark with precise annotations and diverse scenes to facilitate future research.

Rui Zhang, Yang Han, Zilong Huang et al.

This is a short technical report introducing the solution of the Team TCParser for Short-video Face Parsing Track of The 3rd Person in Context (PIC) Workshop and Challenge at CVPR 2021. In this paper, we introduce a strong backbone which is cross-window based Shuffle Transformer for presenting accurate face parsing representation. To further obtain the finer segmentation results, especially on the edges, we introduce a Feature Alignment Aggregation (FAA) module. It can effectively relieve the feature misalignment issue caused by multi-resolution feature aggregation. Benefiting from the stronger backbone and better feature aggregation, the proposed method achieves 86.9519% score in the Short-video Face Parsing track of the 3rd Person in Context (PIC) Workshop and Challenge, ranked the first place.

Zilong Huang, Youcheng Ben, Guozhong Luo et al.

Very recently, Window-based Transformers, which computed self-attention within non-overlapping local windows, demonstrated promising results on image classification, semantic segmentation, and object detection. However, less study has been devoted to the cross-window connection which is the key element to improve the representation ability. In this work, we revisit the spatial shuffle as an efficient way to build connections among windows. As a result, we propose a new vision transformer, named Shuffle Transformer, which is highly efficient and easy to implement by modifying two lines of code. Furthermore, the depth-wise convolution is introduced to complement the spatial shuffle for enhancing neighbor-window connections. The proposed architectures achieve excellent performance on a wide range of visual tasks including image-level classification, object detection, and semantic segmentation. Code will be released for reproduction.

Andrey Ignatov, Grigory Malivenko, David Plowman et al.

Depth estimation is an important computer vision problem with many practical applications to mobile devices. While many solutions have been proposed for this task, they are usually very computationally expensive and thus are not applicable for on-device inference. To address this problem, we introduce the first Mobile AI challenge, where the target is to develop an end-to-end deep learning-based depth estimation solutions that can demonstrate a nearly real-time performance on smartphones and IoT platforms. For this, the participants were provided with a new large-scale dataset containing RGB-depth image pairs obtained with a dedicated stereo ZED camera producing high-resolution depth maps for objects located at up to 50 meters. The runtime of all models was evaluated on the popular Raspberry Pi 4 platform with a mobile ARM-based Broadcom chipset. The proposed solutions can generate VGA resolution depth maps at up to 10 FPS on the Raspberry Pi 4 while achieving high fidelity results, and are compatible with any Android or Linux-based mobile devices. A detailed description of all models developed in the challenge is provided in this paper.

Zilong Huang, Wentian Hao, Xinggang Wang et al.

Despite their success for semantic segmentation, convolutional neural networks are ill-equipped for incremental learning, \ie, adapting the original segmentation model as new classes are available but the initial training data is not retained. Actually, they are vulnerable to catastrophic forgetting problem. We try to address this issue by "inverting" the trained segmentation network to synthesize input images starting from random noise. To avoid setting detailed pixel-wise segmentation maps as the supervision manually, we propose the SegInversion to synthesize images using the image-level labels. To increase the diversity of synthetic images, the Scale-Aware Aggregation module is integrated into SegInversion for controlling the scale (the number of pixels) of synthetic objects. Along with real images of new classes, the synthesized images will be fed into the distillation-based framework to train the new segmentation model which retains the information about previously learned classes, whilst updating the current model to learn the new ones. The proposed method significantly outperforms other incremental learning methods and obtains state-of-the-art performance on the PASCAL VOC 2012 and ADE20K datasets. The code and models will be made publicly available.

Qiang Zhou, Shiyin Wang, Yitong Wang et al.

In this paper, we tackle the problem of human de-occlusion which reasons about occluded segmentation masks and invisible appearance content of humans. In particular, a two-stage framework is proposed to estimate the invisible portions and recover the content inside. For the stage of mask completion, a stacked network structure is devised to refine inaccurate masks from a general instance segmentation model and predict integrated masks simultaneously. Additionally, the guidance from human parsing and typical pose masks are leveraged to bring prior information. For the stage of content recovery, a novel parsing guided attention module is applied to isolate body parts and capture context information across multiple scales. Besides, an Amodal Human Perception dataset (AHP) is collected to settle the task of human de-occlusion. AHP has advantages of providing annotations from real-world scenes and the number of humans is comparatively larger than other amodal perception datasets. Based on this dataset, experiments demonstrate that our method performs over the state-of-the-art techniques in both tasks of mask completion and content recovery. Our AHP dataset is available at \url{https://sydney0zq.github.io/ahp/}.