Publications

This paper's primary objective is to develop a robust generalist perception model capable of addressing multiple tasks under constraints of computational resources and limited training data. We leverage text-to-image diffusion models pre-trained on billions of images and successfully introduce our DICEPTION, a visual generalist model. Exhaustive evaluations demonstrate that DICEPTION effectively tackles diverse perception tasks, even achieving performance comparable to SOTA single-task specialist models. Specifically, we achieve results on par with SAM-vit-h using only 0.06% of their data (e.g., 600K vs. 1B pixel-level annotated images). We designed comprehensive experiments on architectures and input paradigms, demonstrating that the key to successfully re-purposing a single diffusion model for multiple perception tasks lies in maximizing the preservation of the pre-trained model's prior knowledge. Consequently, DICEPTION can be trained with substantially lower computational costs than conventional models requiring training from scratch. Furthermore, adapting DICEPTION to novel tasks is highly efficient, necessitating fine-tuning on as few as 50 images and approximately 1% of its parameters. Finally, we demonstrate that a subtle application of classifier-free guidance can improve the model's performance on depth and normal estimation. We also show that pixel-aligned training, as is characteristic of perception tasks, significantly enhances the model's ability to preserve fine details. DICEPTION offers valuable insights and presents a promising direction for the development of advanced diffusion-based visual generalist models.

Glasses-free three-dimensional (3D) displays provide users with an immersive visual experience without the need of any wearable devices. To achieve high-quality 3D imaging, a display should have both large linear dimensions and a wide viewing angle. However, the trade-off between spatial extent and bandwidth of optical systems, the space–bandwidth product, conventionally constrains the simultaneous maximization of the two. The two most common approaches to 3D displays are holographic and automultiscopic, which, respectively, sacrifice either scale or viewing angle. Recently, some implementations enhanced by artificial intelligence have shown directions to mitigate these constraints, but they still operate within a set space–bandwidth product. As a result, it remains challenging to fabricate large-scale wide-angle 3D displays. Here we report the realization of a large-scale full-parallax 3D display with seamless viewing beyond 100°, maintained at over 50 Hz and 1,920 × 1,080 resolution on a low-cost light-field delivery setup. This device, called EyeReal, is realized by accurately modelling binocular view and combining it with a deep-learning real-time optimization, enabling the generation of optimal light-field outputs for each of the eyes. Our device could potentially enable applications in educational tools, 3D design and virtual reality.

Recent advancements in video generation have primarily leveraged diffusion models for short-duration content. However, these approaches often fall short in modeling complex narratives and maintaining character consistency over extended periods, which is essential for long-form video production like movies. We propose MovieDreamer, a novel hierarchical framework that integrates the strengths of autoregressive models with diffusion-based rendering to pioneer long-duration video generation with intricate plot progressions and high visual fidelity. Our approach utilizes autoregressive models for global narrative coherence, predicting sequences of visual tokens that are subsequently transformed into high-quality video frames through diffusion rendering. This method is akin to traditional movie production processes, where complex stories are factorized down into manageable scene capturing. Further, we employ a multimodal script that enriches scene descriptions with detailed character information and visual style, enhancing continuity and character identity across scenes. We present extensive experiments across various movie genres, demonstrating that our approach not only achieves superior visual and narrative quality but also effectively extends the duration of generated content significantly beyond current capabilities.

Story visualization aims to generate a series of images that match the story described in texts, and it requires the generated images to satisfy high quality, alignment with the text description, and consistency in character identities. Given the complexity of story visualization, existing methods drastically simplify the problem by considering only a few specific characters and scenarios, or requiring the users to provide per-image control conditions such as sketches. However, these simplifications render these methods incompetent for real applications. To this end, we propose an automated story visualization system that can effectively generate diverse, high-quality, and consistent sets of story images, with minimal human interactions. Specifically, we utilize the comprehension and planning capabilities of large language models for layout planning, and then leverage large-scale text-to-image models to generate sophisticated story images based on the layout. We empirically find that sparse control conditions, such as bounding boxes, are suitable for layout planning, while dense control conditions, e.g., sketches, and keypoints, are suitable for generating high-quality image content. To obtain the best of both worlds, we devise a dense condition generation module to transform simple bounding box layouts into sketch or keypoint control conditions for final image generation, which not only improves the image quality but also allows easy and intuitive user interactions. In addition, we propose a simple yet effective method to generate multi-view consistent character images, eliminating the reliance on human labor to collect or draw character images. This allows our method to obtain consistent story visualization even when only texts are provided as input. Both qualitative and quantitative experiments demonstrate the superiority of our method.

Benefiting from large-scale pre-trained text-to-image (T2I) generative models impressive progress has been achieved in customized image generation which aims to generate user-specified concepts. Existing approaches have extensively focused on single-concept customization and still encounter challenges when it comes to complex scenarios that involve combining multiple concepts. These approaches often require retraining/fine-tuning using a few images leading to time-consuming training processes and impeding their swift implementation. Furthermore the reliance on multiple images to represent a singular concept increases the difficulty of customization. To this end we propose FreeCustom a novel tuning-free method to generate customized images of multi-concept composition based on reference concepts using only one image per concept as input. Specifically we introduce a new multi-reference self-attention (MRSA) mechanism and a weighted mask strategy that enables the generated image to access and focus more on the reference concepts. In addition MRSA leverages our key finding that input concepts are better preserved when providing images with context interactions. Experiments show that our method's produced images are consistent with the given concepts and better aligned with the input text. Our method outperforms or performs on par with other training-based methods in terms of multi-concept composition and single-concept customization but is simpler.

Long-horizon video-audio reasoning and fine-grained pixel understanding impose conflicting requirements on omnimodal models: dense temporal coverage demands many low-resolution frames, whereas precise grounding calls for high-resolution inputs. We tackle this trade-off with a two-system architecture: a Global Reasoning System selects informative keyframes and rewrites the task at low spatial cost, while a Detail Understanding System performs pixel-level grounding on the selected high-resolution snippets. Because ``optimal'' keyframe selection and reformulation are ambiguous and hard to supervise, we formulate them as a reinforcement learning (RL) problem and present Omni-R1, an end-to-end RL framework built on Group Relative Policy Optimization. Omni-R1 trains the Global Reasoning System through hierarchical rewards obtained via online collaboration with the Detail Understanding System, requiring only one epoch of RL on small task splits. Experiments on two challenging benchmarks, namely Referring Audio-Visual Segmentation (RefAVS) and Reasoning Video Object Segmentation (REVOS), show that Omni-R1 not only surpasses strong supervised baselines but also outperforms specialized state-of-the-art models, while substantially improving out-of-domain generalization and mitigating multimodal hallucination. Our results demonstrate the first successful application of RL to large-scale omnimodal reasoning and highlight a scalable path toward universally foundation models.

In this paper, we tackle the issue of customized image and video generation, encompassing both single-object and multi-object customization. Current methods primarily concentrate on single-object customization and struggle with more complex scenarios involving multiple concepts. Additionally, these methods often necessitate retraining or fine-tuning with a few examples, leading to a time-consuming process that impedes rapid adaptability. To address these weaknesses, we present FreerCustom, a method that eliminates the need for additional training to generate customized images and videos featuring both single and multi-concept compositions, using only one reference image per concept. Our method features a novel multi-reference self-attention (MRSA) mechanism coupled with a weighted mask strategy, allowing the model to better capture and the desired concepts in the generated images. Additionally, MRSA leverages our critical observation that customized generation is more effective when the images provided have contextual interactions. Experimental results show that our method not only closely aligns the generated results with the target concepts and text inputs, but also performs equally well or even surpasses existing training-based methods in both multi-concept and single-concept scenarios.

Active vision, also known as active perception, refers to the process of actively selecting where and how to look in order to gather task-relevant information. It is a critical component of efficient perception and decision-making in humans and advanced embodied agents. Recently, the use of Multimodal Large Language Models (MLLMs) as central planning and decision-making modules in robotic systems has gained extensive attention. However, despite the importance of active perception in embodied intelligence, there is little to no exploration of how MLLMs can be equipped with or learn active perception capabilities. In this paper, we first provide a systematic definition of MLLM-based active perception tasks. We point out that the recently proposed GPT-o3 model's zoom-in search strategy can be regarded as a special case of active perception; however, it still suffers from low search efficiency and inaccurate region selection. To address these issues, we propose ACTIVE-O3, a purely reinforcement learning based training framework built on top of GRPO, designed to equip MLLMs with active perception capabilities. We further establish a comprehensive benchmark suite to evaluate ACTIVE-O3 across both general open-world tasks, such as small-object and dense object grounding, and domain-specific scenarios, including small object detection in remote sensing and autonomous driving, as well as fine-grained interactive segmentation. In addition, ACTIVE-O3 also demonstrates strong zero-shot reasoning abilities on the V* Benchmark, without relying on any explicit reasoning data. We hope that our work can provide a simple codebase and evaluation protocol to facilitate future research on active perception in MLLMs.

We introduce Tinker, a versatile framework for high-fidelity 3D editing that operates in both one-shot and few-shot regimes without any per-scene finetuning. Unlike prior techniques that demand extensive per-scene optimization to ensure multi-view consistency or to produce dozens of consistent edited input views, Tinker delivers robust, multi-view consistent edits from as few as one or two images. This capability stems from repurposing pretrained diffusion models, which unlocks their latent 3D awareness. To drive research in this space, we curate the first large-scale multi-view editing dataset and data pipeline, spanning diverse scenes and styles. Building on this dataset, we develop our framework capable of generating multi-view consistent edited views without per-scene training, which consists of two novel components: (1) Referring multi-view editor: Enables precise, reference-driven edits that remain coherent across all viewpoints. (2) Any-view-to-video synthesizer: Leverages spatial-temporal priors from video diffusion to perform high-quality scene completion and novel-view generation even from sparse inputs. Through extensive experiments, Tinker significantly reduces the barrier to generalizable 3D content creation, achieving state-of-the-art performance on editing, novel-view synthesis, and rendering enhancement tasks. We believe that Tinker represents a key step towards truly scalable, zero-shot 3D editing.

A fundamental challenge in embodied intelligence is developing expressive and compact state representations for efficient world modeling and decision making. However, existing methods often fail to achieve this balance, yielding representations that are either overly redundant or lacking in task-critical information. We propose an unsupervised approach that learns a highly compressed two-token state representation using a lightweight encoder and a pre-trained Diffusion Transformer (DiT) decoder, capitalizing on its strong generative prior. Our representation is efficient, interpretable, and integrates seamlessly into existing VLA-based models, improving performance by 14.3% on LIBERO and 30% in real-world task success with minimal inference overhead. More importantly, we find that the difference between these tokens, obtained via latent interpolation, naturally serves as a highly effective latent action, which can be further decoded into executable robot actions. This emergent capability reveals that our representation captures structured dynamics without explicit supervision. We name our method StaMo for its ability to learn generalizable robotic Motion from compact State representation, which is encoded from static images, challenging the prevalent dependence to learning latent action on complex architectures and video data. The resulting latent actions also enhance policy co-training, outperforming prior methods by 10.4% with improved interpretability. Moreover, our approach scales effectively across diverse data sources, including real-world robot data, simulation, and human egocentric video.

Vision-Language-Action (VLA) models represent a pivotal advance in embodied intelligence, yet they confront critical barriers to real-world deployment, most notably catastrophic forgetting. This issue stems from their overreliance on continuous action sequences or action chunks, which inadvertently create isolated data silos that disrupt knowledge retention across tasks. To tackle these challenges, we propose the Narrowing of Trajectory VLA (NoTVLA) framework: a novel approach that narrows its focus to sparse trajectories, thereby avoiding the catastrophic forgetting associated with dense trajectory fine-tuning. A key innovation of NoTVLA lies in its trajectory planning strategy: instead of centering on the target object's trajectory, it leverages temporal compression and spatial reasoning pruning specifically for the robot end effector's trajectory. Furthermore, training is conducted using these sparse trajectories rather than dense action trajectories, an optimization that delivers remarkable practical advantages with better performance in zero-shot. In multi-task evaluation scenarios, NoTVLA achieves superior performance and generalization compared to pi0 while operating under two critical constraints: it uses over an order of magnitude less computing power than pi0 and requires no wrist-mounted camera. This design ensures that NoTVLA's operational accuracy closely approximates that of single-task expert models. Crucially, it also preserves the model's inherent language capabilities, enabling zero-shot generalization in specific scenarios, supporting unified model deployment across multiple robot platforms, and fostering a degree of generalization even when perceiving tasks from novel perspectives.

Although Vision-Language Models (VLM) have demonstrated impressive planning and reasoning capabilities, translating these abilities into the physical world introduces significant challenges. Conventional Vision-Language-Action (VLA) models, which integrate reasoning and action into a monolithic architecture, generalize poorly because they are constrained by scarce, narrow-domain data. While recent dual-system approaches attempt to decouple thinking from acting, they are often constrained by semantic ambiguities within the action module. This ambiguity makes large-scale, cross-task training infeasible. Consequently, these systems typically necessitate fine-tuning on newly collected data when deployed to novel environments, and the cooperation mechanism between the two systems remains ill-defined. To address these limitations, we introduce, for the first time, a framework centered around a generalizable action expert. Our approach utilizes sparse 3D trajectories as an intermediate representation, effectively bridging the high-level planning capabilities of the VLM with the low-level physical action module. During the planning phase, the VLM is only required to generate coarse 3D waypoints. These waypoints are then processed by our generalizable action expert, which refines them into dense, executable action sequences by sampling real-time point cloud observations of the environment. To promote training efficiency and robust generalization, we introduce a novel Action Pre-training, Pointcloud Fine-tuning paradigm. Our method combines the broad generalization capabilities of VLMs in visual understanding and planning with the fine-grained, action-level generalization of action expert.