Yuyang Zhao

Henry Hengyuan Zhao, Pichao Wang, Yuyang Zhao et al. · stanford

Pre-trained vision transformers have strong representation benefits to various downstream tasks. Recently, many parameter-efficient fine-tuning (PEFT) methods have been proposed, and their experiments demonstrate that tuning only 1\% extra parameters could surpass full fine-tuning in low-data resource scenarios. However, these methods overlook the task-specific information when fine-tuning diverse downstream tasks. In this paper, we propose a simple yet effective method called "Salient Channel Tuning" (SCT) to leverage the task-specific information by forwarding the model with the task images to select partial channels in a feature map that enables us to tune only 1/8 channels leading to significantly lower parameter costs. Experiments on 19 visual transfer learning downstream tasks demonstrate that our SCT outperforms full fine-tuning on 18 out of 19 tasks by adding only 0.11M parameters of the ViT-B, which is 780$\times$ fewer than its full fine-tuning counterpart. Furthermore, experiments on domain generalization and few-shot classification further demonstrate the effectiveness and generic of our approach. The code is available at https://github.com/showlab/SCT.

Yuyang Zhao, Yicheng Pan, Qiyuan He et al.

Real-time streaming video-to-video editing (V2V) is critical for interactive applications such as live broadcasting and gaming, yet it remains a formidable challenge due to the stringent requirements for temporal consistency and inference throughput. In this paper, we present SANA-Streaming, a system-algorithm co-designed framework for high-resolution, real-time streaming video editing on consumer GPUs, with the following three core designs: (1) Hybrid Diffusion Transformer architecture introduces softmax attention in part of the blocks to improve local modeling capabilities while preserving the efficiency of linear layers. (2) Cycle-Reverse Regularization is a novel training strategy that enforces semantic consistency by predicting source frames from generated content via flow matching, improving temporal consistency without requiring paired long edited videos. (3) Efficient System Co-design combines fused GDN kernels and Mixed-Precision Quantization (MPQ) optimized for the NVIDIA Blackwell (RTX 5090) architecture. By profiling real-world throughput, our MPQ maximizes Tensor Core utilization while maintaining generation quality. The resulting system achieves real-time 1280 x 704 resolution editing at 24 end-to-end FPS on a single RTX 5090 GPU, with the DiT core running at 58 FPS. Experimental results demonstrate that our co-design approach significantly outperforms existing SOTA methods in both temporal coherence and system throughput.

Yuyang Zhao, Zhun Zhong, Na Zhao et al.

In this paper, we study the task of synthetic-to-real domain generalized semantic segmentation, which aims to learn a model that is robust to unseen real-world scenes using only synthetic data. The large domain shift between synthetic and real-world data, including the limited source environmental variations and the large distribution gap between synthetic and real-world data, significantly hinders the model performance on unseen real-world scenes. In this work, we propose the Style-HAllucinated Dual consistEncy learning (SHADE) framework to handle such domain shift. Specifically, SHADE is constructed based on two consistency constraints, Style Consistency (SC) and Retrospection Consistency (RC). SC enriches the source situations and encourages the model to learn consistent representation across style-diversified samples. RC leverages real-world knowledge to prevent the model from overfitting to synthetic data and thus largely keeps the representation consistent between the synthetic and real-world models. Furthermore, we present a novel style hallucination module (SHM) to generate style-diversified samples that are essential to consistency learning. SHM selects basis styles from the source distribution, enabling the model to dynamically generate diverse and realistic samples during training. Experiments show that our SHADE yields significant improvement and outperforms state-of-the-art methods by 5.05% and 8.35% on the average mIoU of three real-world datasets on single- and multi-source settings, respectively.

Hengyuan Zhao, Hao Luo, Yuyang Zhao et al. · stanford

Parameter-Efficient Transfer Learning (PETL) aims at efficiently adapting large models pre-trained on massive data to downstream tasks with limited task-specific data. In view of the practicality of PETL, previous works focus on tuning a small set of parameters for each downstream task in an end-to-end manner while rarely considering the task distribution shift issue between the pre-training task and the downstream task. This paper proposes a novel two-stage paradigm, where the pre-trained model is first aligned to the target distribution. Then the task-relevant information is leveraged for effective adaptation. Specifically, the first stage narrows the task distribution shift by tuning the scale and shift in the LayerNorm layers. In the second stage, to efficiently learn the task-relevant information, we propose a Taylor expansion-based importance score to identify task-relevant channels for the downstream task and then only tune such a small portion of channels, making the adaptation to be parameter-efficient. Overall, we present a promising new direction for PETL, and the proposed paradigm achieves state-of-the-art performance on the average accuracy of 19 downstream tasks.

Zhun Zhong, Yuyang Zhao, Gim Hee Lee et al.

In this paper, we consider the problem of domain generalization in semantic segmentation, which aims to learn a robust model using only labeled synthetic (source) data. The model is expected to perform well on unseen real (target) domains. Our study finds that the image style variation can largely influence the model's performance and the style features can be well represented by the channel-wise mean and standard deviation of images. Inspired by this, we propose a novel adversarial style augmentation (AdvStyle) approach, which can dynamically generate hard stylized images during training and thus can effectively prevent the model from overfitting on the source domain. Specifically, AdvStyle regards the style feature as a learnable parameter and updates it by adversarial training. The learned adversarial style feature is used to construct an adversarial image for robust model training. AdvStyle is easy to implement and can be readily applied to different models. Experiments on two synthetic-to-real semantic segmentation benchmarks demonstrate that AdvStyle can significantly improve the model performance on unseen real domains and show that we can achieve the state of the art. Moreover, AdvStyle can be employed to domain generalized image classification and produces a clear improvement on the considered datasets.

Jianing Qiu, Jian Wu, Hao Wei et al.

We present VisionFM, a foundation model pre-trained with 3.4 million ophthalmic images from 560,457 individuals, covering a broad range of ophthalmic diseases, modalities, imaging devices, and demography. After pre-training, VisionFM provides a foundation to foster multiple ophthalmic artificial intelligence (AI) applications, such as disease screening and diagnosis, disease prognosis, subclassification of disease phenotype, and systemic biomarker and disease prediction, with each application enhanced with expert-level intelligence and accuracy. The generalist intelligence of VisionFM outperformed ophthalmologists with basic and intermediate levels in jointly diagnosing 12 common ophthalmic diseases. Evaluated on a new large-scale ophthalmic disease diagnosis benchmark database, as well as a new large-scale segmentation and detection benchmark database, VisionFM outperformed strong baseline deep neural networks. The ophthalmic image representations learned by VisionFM exhibited noteworthy explainability, and demonstrated strong generalizability to new ophthalmic modalities, disease spectrum, and imaging devices. As a foundation model, VisionFM has a large capacity to learn from diverse ophthalmic imaging data and disparate datasets. To be commensurate with this capacity, in addition to the real data used for pre-training, we also generated and leveraged synthetic ophthalmic imaging data. Experimental results revealed that synthetic data that passed visual Turing tests, can also enhance the representation learning capability of VisionFM, leading to substantial performance gains on downstream ophthalmic AI tasks. Beyond the ophthalmic AI applications developed, validated, and demonstrated in this work, substantial further applications can be achieved in an efficient and cost-effective manner using VisionFM as the foundation.

Yuyang Zhao, Zhiwen Yan, Enze Xie et al.

We introduce Animate124 (Animate-one-image-to-4D), the first work to animate a single in-the-wild image into 3D video through textual motion descriptions, an underexplored problem with significant applications. Our 4D generation leverages an advanced 4D grid dynamic Neural Radiance Field (NeRF) model, optimized in three distinct stages using multiple diffusion priors. Initially, a static model is optimized using the reference image, guided by 2D and 3D diffusion priors, which serves as the initialization for the dynamic NeRF. Subsequently, a video diffusion model is employed to learn the motion specific to the subject. However, the object in the 3D videos tends to drift away from the reference image over time. This drift is mainly due to the misalignment between the text prompt and the reference image in the video diffusion model. In the final stage, a personalized diffusion prior is therefore utilized to address the semantic drift. As the pioneering image-text-to-4D generation framework, our method demonstrates significant advancements over existing baselines, evidenced by comprehensive quantitative and qualitative assessments.

Yuyang Zhao, Zhun Zhong, Na Zhao et al.

Domain shift widely exists in the visual world, while modern deep neural networks commonly suffer from severe performance degradation under domain shift due to the poor generalization ability, which limits the real-world applications. The domain shift mainly lies in the limited source environmental variations and the large distribution gap between source and unseen target data. To this end, we propose a unified framework, Style-HAllucinated Dual consistEncy learning (SHADE), to handle such domain shift in various visual tasks. Specifically, SHADE is constructed based on two consistency constraints, Style Consistency (SC) and Retrospection Consistency (RC). SC enriches the source situations and encourages the model to learn consistent representation across style-diversified samples. RC leverages general visual knowledge to prevent the model from overfitting to source data and thus largely keeps the representation consistent between the source and general visual models. Furthermore, we present a novel style hallucination module (SHM) to generate style-diversified samples that are essential to consistency learning. SHM selects basis styles from the source distribution, enabling the model to dynamically generate diverse and realistic samples during training. Extensive experiments demonstrate that our versatile SHADE can significantly enhance the generalization in various visual recognition tasks, including image classification, semantic segmentation and object detection, with different models, i.e., ConvNets and Transformer.

Haoyi Zhu, Haozhe Liu, Yuyang Zhao et al.

We introduce SANA-WM, an efficient 2.6B-parameter open-source world model natively trained for one-minute generation, synthesizing high-fidelity, 720p, minute-scale videos with precise camera control. SANA-WM achieves visual quality comparable to large-scale industrial baselines such as LingBot-World and HY-WorldPlay, while significantly improving efficiency. Four core designs drive our architecture: (1) Hybrid Linear Attention combines frame-wise Gated DeltaNet (GDN) with softmax attention for memory-efficient long-context modeling. (2) Dual-Branch Camera Control ensures precise 6-DoF trajectory adherence. (3) Two-Stage Generation Pipeline applies a long-video refiner to stage-1 outputs, improving quality and consistency across sequences. (4) Robust Annotation Pipeline extracts accurate metric-scale 6-DoF camera poses from public videos to yield high-quality, spatiotemporally consistent action labels. Driven by these designs, SANA-WMdemonstrates remarkable efficiency across data, training compute, and inference hardware: it uses only $\sim$213K public video clips with metric-scale pose supervision, completes training in 15 days on 64 H100s, and generates each 60s clip on a single GPU; its distilled variant can be deployed on a single RTX 5090 with NVFP4 quantization to denoise a 60s 720p clip in 34s. On our one-minute world-model benchmark, SANA-WM demonstrates stronger action-following accuracy than prior open-source baselines and achieves comparable visual quality at $36\times$ higher throughput for scalable world modeling.

Yuyang Zhao, Na Zhao, Gim Hee Lee

Semantic segmentation in 3D indoor scenes has achieved remarkable performance under the supervision of large-scale annotated data. However, previous works rely on the assumption that the training and testing data are of the same distribution, which may suffer from performance degradation when evaluated on the out-of-distribution scenes. To alleviate the annotation cost and the performance degradation, this paper introduces the synthetic-to-real domain generalization setting to this task. Specifically, the domain gap between synthetic and real-world point cloud data mainly lies in the different layouts and point patterns. To address these problems, we first propose a clustering instance mix (CINMix) augmentation technique to diversify the layouts of the source data. In addition, we augment the point patterns of the source data and introduce non-parametric multi-prototypes to ameliorate the intra-class variance enlarged by the augmented point patterns. The multi-prototypes can model the intra-class variance and rectify the global classifier in both training and inference stages. Experiments on the synthetic-to-real benchmark demonstrate that both CINMix and multi-prototypes can narrow the distribution gap and thus improve the generalization ability on real-world datasets.

Enze Xie, Junsong Chen, Yuyang Zhao et al.

This paper presents SANA-1.5, a linear Diffusion Transformer for efficient scaling in text-to-image generation. Building upon SANA-1.0, we introduce three key innovations: (1) Efficient Training Scaling: A depth-growth paradigm that enables scaling from 1.6B to 4.8B parameters with significantly reduced computational resources, combined with a memory-efficient 8-bit optimizer. (2) Model Depth Pruning: A block importance analysis technique for efficient model compression to arbitrary sizes with minimal quality loss. (3) Inference-time Scaling: A repeated sampling strategy that trades computation for model capacity, enabling smaller models to match larger model quality at inference time. Through these strategies, SANA-1.5 achieves a text-image alignment score of 0.81 on GenEval, which can be further improved to 0.96 through inference scaling with VILA-Judge, establishing a new SoTA on GenEval benchmark. These innovations enable efficient model scaling across different compute budgets while maintaining high quality, making high-quality image generation more accessible. Our code and pre-trained models are released.

Wentao Shi, Yu Wang, Yuyang Zhao et al.

As reinforcement learning continues to scale the training of large language model-based agents, reliably verifying agent behaviors in complex environments has become increasingly challenging. Existing approaches rely on rule-based verifiers or LLM-as-a-Judge models, which struggle to generalize beyond narrow domains. Agent-as-a-Judge addresses this limitation by actively interacting with environments and tools to acquire verifiable evidence, yet its capabilities remain underexplored. We introduce a benchmark AJ-Bench to systematically evaluate Agent-as-a-Judge across three domains-search, data systems, and graphical user interfaces-comprising 155 tasks and 516 annotated trajectories. The benchmark comprehensively assesses judge agents' abilities in information acquisition, state verification, and process verification. Experiments demonstrate consistent performance gains over LLM-as-a-Judge baselines, while also revealing substantial open challenges in agent-based verification. Our data and code are available at https://aj-bench.github.io/.

Jiaxun Sun, Hengyu Xue, Yuyang Zhao

We derive a computable closed-form upper bound on the Hausdorff distance between a truncated minimal robust positively invariant (mRPI) set and its infinite-horizon limit. The bound depends only on a disturbance-set size measure and an induced-norm contraction factor of the system matrix, and it yields an explicit, fully analytic horizon-selection rule that guarantees a prescribed approximation tolerance without iterative set computations. The choice of vector norm enters as a design lever: norm shaping -- through diagonal or Lyapunov-based weighting -- tightens both the contraction factor and the resulting certificate, with direct consequences for robust invariant-set approximation and tube-based model predictive control (MPC) constraint tightening. Numerical examples illustrate the accuracy, scalability, and practical impact of the proposed bound.

Mengqi Guo, Chen Li, Yuyang Zhao et al.

Inferring step-wise actions to assemble 3D objects with primitive bricks from images is a challenging task due to complex constraints and the vast number of possible combinations. Recent studies have demonstrated promising results on sequential LEGO brick assembly through the utilization of LEGO-Graph modeling to predict sequential actions. However, existing approaches are class-specific and require significant computational and 3D annotation resources. In this work, we first propose a computationally efficient breadth-first search (BFS) LEGO-Tree structure to model the sequential assembly actions by considering connections between consecutive layers. Based on the LEGO-Tree structure, we then design a class-agnostic tree-transformer framework to predict the sequential assembly actions from the input multi-view images. A major challenge of the sequential brick assembly task is that the step-wise action labels are costly and tedious to obtain in practice. We mitigate this problem by leveraging synthetic-to-real transfer learning. Specifically, our model is first pre-trained on synthetic data with full supervision from the available action labels. We then circumvent the requirement for action labels in the real data by proposing an action-to-silhouette projection that replaces action labels with input image silhouettes for self-supervision. Without any annotation on the real data, our model outperforms existing methods with 3D supervision by 7.8% and 11.3% in mIoU on the MNIST and ModelNet Construction datasets, respectively.

Yukang Chen, Luozhou Wang, Wei Huang et al.

We present LongLive-2.0, an NVFP4-based parallel infrastructure throughout the full training and inference workflow of long video generation, addressing speed and memory bottlenecks. For training, we introduce sequence-parallel autoregressive (AR) training, instantiated as Balanced SP, which co-designs the efficient teacher-forcing layout with SP execution by pairing clean-history and noisy-target temporal chunks on each rank, enabling a natural teacher-forcing mask with SP-aware chunked VAE encoding. Combined with NVFP4 precision, it reduces GPU memory cost and accelerates GEMM computation during training, the proportion of which increases as video length grows. Moreover, we show that a high-quality infrastructure and dataset enable a remarkably clean training pipeline. Unlike existing Self-Forcing series methods that rely on ODE initialization and subsequent distribution matching distillation (DMD), LongLive-2.0 directly tunes a diffusion model into a long, multi-shot, interactive auto-regressive (AR) diffusion model. It can be further converted to real-time generation (4 to 2 denoising steps) with standalone LoRA weights. For inference on Blackwell GPUs, we enable W4A4 NVFP4 inference, quantize KV cache into NVFP4 for memory savings, and boost end-to-end throughput with asynchronous streaming VAE decoding. On non-Blackwell GPU architectures, we deploy SP inference to match the speed on Blackwell GPUs, while the quantized KV cache can lower inter-GPU communication of SP. Experiments show up to 2.15x speedup in training, and 1.84x in inference. LongLive-2.0-5B achieves 45.7 FPS inference while attaining strong performance on benchmarks. To our knowledge, LongLive-2.0 is the first NVFP4 training and inference system for long video generation.

Dongyue Lu, Ao Liang, Tianxin Huang et al.

Immersive applications call for synthesizing spatiotemporal 4D content from casual videos without costly 3D supervision. Existing video-to-4D methods typically rely on manually annotated camera poses, which are labor-intensive and brittle for in-the-wild footage. Recent warp-then-inpaint approaches mitigate the need for pose labels by warping input frames along a novel camera trajectory and using an inpainting model to fill missing regions, thereby depicting the 4D scene from diverse viewpoints. However, this trajectory-to-trajectory formulation often entangles camera motion with scene dynamics and complicates both modeling and inference. We introduce SEE4D, a pose-free, trajectory-to-camera framework that replaces explicit trajectory prediction with rendering to a bank of fixed virtual cameras, thereby separating camera control from scene modeling. A view-conditional video inpainting model is trained to learn a robust geometry prior by denoising realistically synthesized warped images and to inpaint occluded or missing regions across virtual viewpoints, eliminating the need for explicit 3D annotations. Building on this inpainting core, we design a spatiotemporal autoregressive inference pipeline that traverses virtual-camera splines and extends videos with overlapping windows, enabling coherent generation at bounded per-step complexity. We validate See4D on cross-view video generation and sparse reconstruction benchmarks. Across quantitative metrics and qualitative assessments, our method achieves superior generalization and improved performance relative to pose- or trajectory-conditioned baselines, advancing practical 4D world modeling from casual videos.

Junyu Chen, Wenkun He, Yuchao Gu et al.

We introduce DC-VideoGen, a post-training acceleration framework for efficient video generation. DC-VideoGen can be applied to any pre-trained video diffusion model, improving efficiency by adapting it to a deep compression latent space with lightweight fine-tuning. The framework builds on two key innovations: (i) a Deep Compression Video Autoencoder with a novel chunk-causal temporal design that achieves 32x/64x spatial and 4x temporal compression while preserving reconstruction quality and generalization to longer videos; and (ii) AE-Adapt-V, a robust adaptation strategy that enables rapid and stable transfer of pre-trained models into the new latent space. Adapting the pre-trained Wan-2.1-14B model with DC-VideoGen requires only 10 GPU days on the NVIDIA H100 GPU. The accelerated models achieve up to 14.8x lower inference latency than their base counterparts without compromising quality, and further enable 2160x3840 video generation on a single GPU. Code: https://github.com/dc-ai-projects/DC-VideoGen.

Yuyang Zhao, Wentao Shi, Fuli Feng et al.

Large language model (LLM)-based agents have demonstrated remarkable capabilities in addressing complex tasks, thereby enabling more advanced information retrieval and supporting deeper, more sophisticated human information-seeking behaviors. However, most existing agents operate in a purely reactive manner, responding passively to user instructions, which significantly constrains their effectiveness and efficiency as general-purpose platforms for information acquisition. To overcome this limitation, this paper proposes AppAgent-Pro, a proactive GUI agent system that actively integrates multi-domain information based on user instructions. This approach enables the system to proactively anticipate users' underlying needs and conduct in-depth multi-domain information mining, thereby facilitating the acquisition of more comprehensive and intelligent information. AppAgent-Pro has the potential to fundamentally redefine information acquisition in daily life, leading to a profound impact on human society. Our code is available at: https://github.com/LaoKuiZe/AppAgent-Pro. The demonstration video could be found at: https://www.dropbox.com/scl/fi/hvzqo5vnusg66srydzixo/AppAgent-Pro-demo-video.mp4?rlkey=o2nlfqgq6ihl125mcqg7bpgqu&st=d29vrzii&dl=0.

Yuyang Zhao, Chung-Ching Lin, Kevin Lin et al. · microsoft-research

Recent developments in 2D visual generation have been remarkably successful. However, 3D and 4D generation remain challenging in real-world applications due to the lack of large-scale 4D data and effective model design. In this paper, we propose to jointly investigate general 3D and 4D generation by leveraging camera and object movements commonly observed in daily life. Due to the lack of real-world 4D data in the community, we first propose a data curation pipeline to obtain camera poses and object motion strength from videos. Based on this pipeline, we introduce a large-scale real-world 4D scene dataset: CamVid-30K. By leveraging all the 3D and 4D data, we develop our framework, GenXD, which allows us to produce any 3D or 4D scene. We propose multiview-temporal modules, which disentangle camera and object movements, to seamlessly learn from both 3D and 4D data. Additionally, GenXD employs masked latent conditions to support a variety of conditioning views. GenXD can generate videos that follow the camera trajectory as well as consistent 3D views that can be lifted into 3D representations. We perform extensive evaluations across various real-world and synthetic datasets, demonstrating GenXD's effectiveness and versatility compared to previous methods in 3D and 4D generation.

Tao Hu, Wenhang Ge, Yuyang Zhao et al.

We introduce X-Ray, a novel 3D sequential representation inspired by the penetrability of x-ray scans. X-Ray transforms a 3D object into a series of surface frames at different layers, making it suitable for generating 3D models from images. Our method utilizes ray casting from the camera center to capture geometric and textured details, including depth, normal, and color, across all intersected surfaces. This process efficiently condenses the whole 3D object into a multi-frame video format, motivating the utilize of a network architecture similar to those in video diffusion models. This design ensures an efficient 3D representation by focusing solely on surface information. Also, we propose a two-stage pipeline to generate 3D objects from X-Ray Diffusion Model and Upsampler. We demonstrate the practicality and adaptability of our X-Ray representation by synthesizing the complete visible and hidden surfaces of a 3D object from a single input image. Experimental results reveal the state-of-the-art superiority of our representation in enhancing the accuracy of 3D generation, paving the way for new 3D representation research and practical applications.

Junsong Chen, Yuyang Zhao, Jincheng Yu et al.

We introduce SANA-Video, a small diffusion model that can efficiently generate videos up to 720x1280 resolution and minute-length duration. SANA-Video synthesizes high-resolution, high-quality and long videos with strong text-video alignment at a remarkably fast speed, deployable on RTX 5090 GPU. Two core designs ensure our efficient, effective and long video generation: (1) Linear DiT: We leverage linear attention as the core operation, which is more efficient than vanilla attention given the large number of tokens processed in video generation. (2) Constant-Memory KV cache for Block Linear Attention: we design block-wise autoregressive approach for long video generation by employing a constant-memory state, derived from the cumulative properties of linear attention. This KV cache provides the Linear DiT with global context at a fixed memory cost, eliminating the need for a traditional KV cache and enabling efficient, minute-long video generation. In addition, we explore effective data filters and model training strategies, narrowing the training cost to 12 days on 64 H100 GPUs, which is only 1% of the cost of MovieGen. Given its low cost, SANA-Video achieves competitive performance compared to modern state-of-the-art small diffusion models (e.g., Wan 2.1-1.3B and SkyReel-V2-1.3B) while being 16x faster in measured latency. Moreover, SANA-Video can be deployed on RTX 5090 GPUs with NVFP4 precision, accelerating the inference speed of generating a 5-second 720p video from 71s to 29s (2.4x speedup). In summary, SANA-Video enables low-cost, high-quality video generation.

Yiyang Zhou, Linjie Li, Shi Qiu et al. · microsoft-research

Existing video benchmarks often resemble image-based benchmarks, with question types like "What actions does the person perform throughout the video?" or "What color is the woman's dress in the video?" For these, models can often answer by scanning just a few key frames, without deep temporal reasoning. This limits our ability to assess whether large vision-language models (LVLMs) can truly think with videos rather than perform superficial frame-level analysis. To address this, we introduce GLIMPSE, a benchmark specifically designed to evaluate whether LVLMs can genuinely think with videos. Unlike prior benchmarks, GLIMPSE emphasizes comprehensive video understanding beyond static image cues. It consists of 3,269 videos and over 4,342 highly visual-centric questions across 11 categories, including Trajectory Analysis, Temporal Reasoning, and Forensics Detection. All questions are carefully crafted by human annotators and require watching the entire video and reasoning over full video context-this is what we mean by thinking with video. These questions cannot be answered by scanning selected frames or relying on text alone. In human evaluations, GLIMPSE achieves 94.82% accuracy, but current LVLMs face significant challenges. Even the best-performing model, GPT-o3, reaches only 66.43%, highlighting that LVLMs still struggle to move beyond surface-level reasoning to truly think with videos.

Seungjun Lee, Yuyang Zhao, Gim Hee Lee

In this paper, we investigate Open-Vocabulary 3D Instance Segmentation (OV-3DIS) with free-form language instructions. Earlier works that rely on only annotated base categories for training suffer from limited generalization to unseen novel categories. Recent works mitigate poor generalizability to novel categories by generating class-agnostic masks or projecting generalized masks from 2D to 3D, but disregard semantic or geometry information, leading to sub-optimal performance. Instead, generating generalizable but semantic-related masks directly from 3D point clouds would result in superior outcomes. In this paper, we introduce Segment any 3D Object with LanguagE (SOLE), which is a semantic and geometric-aware visual-language learning framework with strong generalizability by generating semantic-related masks directly from 3D point clouds. Specifically, we propose a multimodal fusion network to incorporate multimodal semantics in both backbone and decoder. In addition, to align the 3D segmentation model with various language instructions and enhance the mask quality, we introduce three types of multimodal associations as supervision. Our SOLE outperforms previous methods by a large margin on ScanNetv2, ScanNet200, and Replica benchmarks, and the results are even close to the fully-supervised counterpart despite the absence of class annotations in the training. Furthermore, extensive qualitative results demonstrate the versatility of our SOLE to language instructions.

Shuai Yang, Wei Huang, Ruihang Chu et al.

We present LongLive, a frame-level autoregressive (AR) framework for real-time and interactive long video generation. Long video generation presents challenges in both efficiency and quality. Diffusion and Diffusion-Forcing models can produce high-quality videos but suffer from low efficiency due to bidirectional attention. Causal attention AR models support KV caching for faster inference, but often degrade in quality on long videos due to memory challenges during long-video training. In addition, beyond static prompt-based generation, interactive capabilities, such as streaming prompt inputs, are critical for dynamic content creation, enabling users to guide narratives in real time. This interactive requirement significantly increases complexity, especially in ensuring visual consistency and semantic coherence during prompt transitions. To address these challenges, LongLive adopts a causal, frame-level AR design that integrates a KV-recache mechanism that refreshes cached states with new prompts for smooth, adherent switches; streaming long tuning to enable long video training and to align training and inference (train-long-test-long); and short window attention paired with a frame-level attention sink, shorten as frame sink, preserving long-range consistency while enabling faster generation. With these key designs, LongLive fine-tunes a 1.3B-parameter short-clip model to minute-long generation in just 32 GPU-days. At inference, LongLive sustains 20.7 FPS on a single NVIDIA H100, achieves strong performance on VBench in both short and long videos. LongLive supports up to 240-second videos on a single H100 GPU. LongLive further supports INT8-quantized inference with only marginal quality loss.

Tianxin Huang, Zhiwen Yan, Yuyang Zhao et al.

3D point clouds directly collected from objects through sensors are often incomplete due to self-occlusion. Conventional methods for completing these partial point clouds rely on manually organized training sets and are usually limited to object categories seen during training. In this work, we propose a test-time framework for completing partial point clouds across unseen categories without any requirement for training. Leveraging point rendering via Gaussian Splatting, we develop techniques of Partial Gaussian Initialization, Zero-shot Fractal Completion, and Point Cloud Extraction that utilize priors from pre-trained 2D diffusion models to infer missing regions and extract uniform completed point clouds. Experimental results on both synthetic and real-world scanned point clouds demonstrate that our approach outperforms existing methods in completing a variety of objects. Our project page is at \url{https://tianxinhuang.github.io/projects/ComPC/}.

Yuyang Zhao, Enze Xie, Lanqing Hong et al.

The text-driven image and video diffusion models have achieved unprecedented success in generating realistic and diverse content. Recently, the editing and variation of existing images and videos in diffusion-based generative models have garnered significant attention. However, previous works are limited to editing content with text or providing coarse personalization using a single visual clue, rendering them unsuitable for indescribable content that requires fine-grained and detailed control. In this regard, we propose a generic video editing framework called Make-A-Protagonist, which utilizes textual and visual clues to edit videos with the goal of empowering individuals to become the protagonists. Specifically, we leverage multiple experts to parse source video, target visual and textual clues, and propose a visual-textual-based video generation model that employs mask-guided denoising sampling to generate the desired output. Extensive results demonstrate the versatile and remarkable editing capabilities of Make-A-Protagonist.

Yuyang Zhao, Zhun Zhong, Nicu Sebe et al.

We introduce a new setting of Novel Class Discovery in Semantic Segmentation (NCDSS), which aims at segmenting unlabeled images containing new classes given prior knowledge from a labeled set of disjoint classes. In contrast to existing approaches that look at novel class discovery in image classification, we focus on the more challenging semantic segmentation. In NCDSS, we need to distinguish the objects and background, and to handle the existence of multiple classes within an image, which increases the difficulty in using the unlabeled data. To tackle this new setting, we leverage the labeled base data and a saliency model to coarsely cluster novel classes for model training in our basic framework. Additionally, we propose the Entropy-based Uncertainty Modeling and Self-training (EUMS) framework to overcome noisy pseudo-labels, further improving the model performance on the novel classes. Our EUMS utilizes an entropy ranking technique and a dynamic reassignment to distill clean labels, thereby making full use of the noisy data via self-supervised learning. We build the NCDSS benchmark on the PASCAL-5$^i$ dataset and COCO-20$^i$ dataset. Extensive experiments demonstrate the feasibility of the basic framework (achieving an average mIoU of 49.81% on PASCAL-5$^i$) and the effectiveness of EUMS framework (outperforming the basic framework by 9.28% mIoU on PASCAL-5$^i$).

Kai Liu, Yuyang Zhao, Gang Wang et al.

Cooperative problems under continuous control have always been the focus of multi-agent reinforcement learning. Existing algorithms suffer from the problem of uneven learning degree with the increase of the number of agents. In this paper, a new structure for a multi-agent actor critic is proposed, and the self-attention mechanism is applied in the critic network and the value decomposition method used to solve the uneven problem. The proposed algorithm makes full use of the samples in the replay memory buffer to learn the behavior of a class of agents. First, a new update method is proposed for policy networks that promotes learning efficiency. Second, the utilization of samples is improved, at the same time reflecting the ability of perspective-taking among groups. Finally, the "deceptive signal" in training is eliminated and the learning degree among agents is more uniform than in the existing methods. Multiple experiments were conducted in two typical scenarios of a multi-agent particle environment. Experimental results show that the proposed algorithm can perform better than the state-of-the-art ones, and that it exhibits higher learning efficiency with an increasing number of agents.

Yuyang Zhao, Zhun Zhong, Zhiming Luo et al.

In this work, we introduce a new concept, named source-free open compound domain adaptation (SF-OCDA), and study it in semantic segmentation. SF-OCDA is more challenging than the traditional domain adaptation but it is more practical. It jointly considers (1) the issues of data privacy and data storage and (2) the scenario of multiple target domains and unseen open domains. In SF-OCDA, only the source pre-trained model and the target data are available to learn the target model. The model is evaluated on the samples from the target and unseen open domains. To solve this problem, we present an effective framework by separating the training process into two stages: (1) pre-training a generalized source model and (2) adapting a target model with self-supervised learning. In our framework, we propose the Cross-Patch Style Swap (CPSS) to diversify samples with various patch styles in the feature-level, which can benefit the training of both stages. First, CPSS can significantly improve the generalization ability of the source model, providing more accurate pseudo-labels for the latter stage. Second, CPSS can reduce the influence of noisy pseudo-labels and also avoid the model overfitting to the target domain during self-supervised learning, consistently boosting the performance on the target and open domains. Experiments demonstrate that our method produces state-of-the-art results on the C-Driving dataset. Furthermore, our model also achieves the leading performance on CityScapes for domain generalization.

Yuyang Zhao, Zhun Zhong, Fengxiang Yang et al.

Recent advances in person re-identification (ReID) obtain impressive accuracy in the supervised and unsupervised learning settings. However, most of the existing methods need to train a new model for a new domain by accessing data. Due to public privacy, the new domain data are not always accessible, leading to a limited applicability of these methods. In this paper, we study the problem of multi-source domain generalization in ReID, which aims to learn a model that can perform well on unseen domains with only several labeled source domains. To address this problem, we propose the Memory-based Multi-Source Meta-Learning (M$^3$L) framework to train a generalizable model for unseen domains. Specifically, a meta-learning strategy is introduced to simulate the train-test process of domain generalization for learning more generalizable models. To overcome the unstable meta-optimization caused by the parametric classifier, we propose a memory-based identification loss that is non-parametric and harmonizes with meta-learning. We also present a meta batch normalization layer (MetaBN) to diversify meta-test features, further establishing the advantage of meta-learning. Experiments demonstrate that our M$^3$L can effectively enhance the generalization ability of the model for unseen domains and can outperform the state-of-the-art methods on four large-scale ReID datasets.

Xiaolin Song, Yuyang Zhao, Jingyu Yang

In this paper, we propose a spatio-temporal contextual network, STC-Flow, for optical flow estimation. Unlike previous optical flow estimation approaches with local pyramid feature extraction and multi-level correlation, we propose a contextual relation exploration architecture by capturing rich long-range dependencies in spatial and temporal dimensions. Specifically, STC-Flow contains three key context modules - pyramidal spatial context module, temporal context correlation module and recurrent residual contextual upsampling module, to build the relationship in each stage of feature extraction, correlation, and flow reconstruction, respectively. Experimental results indicate that the proposed scheme achieves the state-of-the-art performance of two-frame based methods on the Sintel dataset and the KITTI 2012/2015 datasets.

Xiaolin Song, Yuyang Zhao, Jingyu Yang et al.

Optical flow estimation is an important yet challenging problem in the field of video analytics. The features of different semantics levels/layers of a convolutional neural network can provide information of different granularity. To exploit such flexible and comprehensive information, we propose a semi-supervised Feature Pyramidal Correlation and Residual Reconstruction Network (FPCR-Net) for optical flow estimation from frame pairs. It consists of two main modules: pyramid correlation mapping and residual reconstruction. The pyramid correlation mapping module takes advantage of the multi-scale correlations of global/local patches by aggregating features of different scales to form a multi-level cost volume. The residual reconstruction module aims to reconstruct the sub-band high-frequency residuals of finer optical flow in each stage. Based on the pyramid correlation mapping, we further propose a correlation-warping-normalization (CWN) module to efficiently exploit the correlation dependency. Experiment results show that the proposed scheme achieves the state-of-the-art performance, with improvement by 0.80, 1.15 and 0.10 in terms of average end-point error (AEE) against competing baseline methods - FlowNet2, LiteFlowNet and PWC-Net on the Final pass of Sintel dataset, respectively.