Lu Jiang

Huiwen Chang, Han Zhang, Jarred Barber et al. · deepmind

We present Muse, a text-to-image Transformer model that achieves state-of-the-art image generation performance while being significantly more efficient than diffusion or autoregressive models. Muse is trained on a masked modeling task in discrete token space: given the text embedding extracted from a pre-trained large language model (LLM), Muse is trained to predict randomly masked image tokens. Compared to pixel-space diffusion models, such as Imagen and DALL-E 2, Muse is significantly more efficient due to the use of discrete tokens and requiring fewer sampling iterations; compared to autoregressive models, such as Parti, Muse is more efficient due to the use of parallel decoding. The use of a pre-trained LLM enables fine-grained language understanding, translating to high-fidelity image generation and the understanding of visual concepts such as objects, their spatial relationships, pose, cardinality etc. Our 900M parameter model achieves a new SOTA on CC3M, with an FID score of 6.06. The Muse 3B parameter model achieves an FID of 7.88 on zero-shot COCO evaluation, along with a CLIP score of 0.32. Muse also directly enables a number of image editing applications without the need to fine-tune or invert the model: inpainting, outpainting, and mask-free editing. More results are available at https://muse-model.github.io

Liangzhe Yuan, Nitesh Bharadwaj Gundavarapu, Long Zhao et al. · deepmind

We evaluate the video understanding capabilities of existing foundation models (FMs) using a carefully designed experiment protocol consisting of three hallmark tasks (action recognition,temporal localization, and spatiotemporal localization), eight datasets well received by the community, and four adaptation methods tailoring an FM for downstream tasks. Furthermore, we jointly profile FMs' efficacy and efficiency when adapting to general video understanding tasks using cost measurements during both training and inference. Our main findings areas follows. First, task-specialized models significantly outperform the seven FMs studied in this work, in sharp contrast to what FMs have achieved in natural language and image understanding. Second, video-native FMs, whose pretraining data mainly contains the video modality, are generally better than image-native FMs in classifying motion-rich videos, localizing actions in time, and understanding a video of more than one action. Third, the video-native FMs can perform well on video tasks under light adaptations to downstream tasks (e.g., freezing the FM backbones), while image-native FMs win in full end-to-end finetuning. The first two observations reveal the need and tremendous opportunities to conduct research on video-focused FMs, and the last confirms that both tasks and adaptation methods matter when it comes to the evaluation of FMs. Our code is released under: https://github.com/tensorflow/models/tree/master/official/projects/videoglue.

Lijun Yu, Yong Cheng, Kihyuk Sohn et al. · cmu, deepmind

We introduce the MAsked Generative VIdeo Transformer, MAGVIT, to tackle various video synthesis tasks with a single model. We introduce a 3D tokenizer to quantize a video into spatial-temporal visual tokens and propose an embedding method for masked video token modeling to facilitate multi-task learning. We conduct extensive experiments to demonstrate the quality, efficiency, and flexibility of MAGVIT. Our experiments show that (i) MAGVIT performs favorably against state-of-the-art approaches and establishes the best-published FVD on three video generation benchmarks, including the challenging Kinetics-600. (ii) MAGVIT outperforms existing methods in inference time by two orders of magnitude against diffusion models and by 60x against autoregressive models. (iii) A single MAGVIT model supports ten diverse generation tasks and generalizes across videos from different visual domains. The source code and trained models will be released to the public at https://magvit.cs.cmu.edu.

Lijun Yu, José Lezama, Nitesh B. Gundavarapu et al. · cmu, deepmind

While Large Language Models (LLMs) are the dominant models for generative tasks in language, they do not perform as well as diffusion models on image and video generation. To effectively use LLMs for visual generation, one crucial component is the visual tokenizer that maps pixel-space inputs to discrete tokens appropriate for LLM learning. In this paper, we introduce MAGVIT-v2, a video tokenizer designed to generate concise and expressive tokens for both videos and images using a common token vocabulary. Equipped with this new tokenizer, we show that LLMs outperform diffusion models on standard image and video generation benchmarks including ImageNet and Kinetics. In addition, we demonstrate that our tokenizer surpasses the previously top-performing video tokenizer on two more tasks: (1) video compression comparable to the next-generation video codec (VCC) according to human evaluations, and (2) learning effective representations for action recognition tasks.

Kihyuk Sohn, Nataniel Ruiz, Kimin Lee et al. · deepmind

Pre-trained large text-to-image models synthesize impressive images with an appropriate use of text prompts. However, ambiguities inherent in natural language and out-of-distribution effects make it hard to synthesize image styles, that leverage a specific design pattern, texture or material. In this paper, we introduce StyleDrop, a method that enables the synthesis of images that faithfully follow a specific style using a text-to-image model. The proposed method is extremely versatile and captures nuances and details of a user-provided style, such as color schemes, shading, design patterns, and local and global effects. It efficiently learns a new style by fine-tuning very few trainable parameters (less than $1\%$ of total model parameters) and improving the quality via iterative training with either human or automated feedback. Better yet, StyleDrop is able to deliver impressive results even when the user supplies only a single image that specifies the desired style. An extensive study shows that, for the task of style tuning text-to-image models, StyleDrop implemented on Muse convincingly outperforms other methods, including DreamBooth and textual inversion on Imagen or Stable Diffusion. More results are available at our project website: https://styledrop.github.io

Kihyuk Sohn, Yuan Hao, José Lezama et al. · deepmind

Transferring knowledge from an image synthesis model trained on a large dataset is a promising direction for learning generative image models from various domains efficiently. While previous works have studied GAN models, we present a recipe for learning vision transformers by generative knowledge transfer. We base our framework on state-of-the-art generative vision transformers that represent an image as a sequence of visual tokens to the autoregressive or non-autoregressive transformers. To adapt to a new domain, we employ prompt tuning, which prepends learnable tokens called prompt to the image token sequence, and introduce a new prompt design for our task. We study on a variety of visual domains, including visual task adaptation benchmark~\cite{zhai2019large}, with varying amount of training images, and show effectiveness of knowledge transfer and a significantly better image generation quality over existing works.

Lijun Yu, Yong Cheng, Zhiruo Wang et al. · cmu, deepmind

In this work, we introduce Semantic Pyramid AutoEncoder (SPAE) for enabling frozen LLMs to perform both understanding and generation tasks involving non-linguistic modalities such as images or videos. SPAE converts between raw pixels and interpretable lexical tokens (or words) extracted from the LLM's vocabulary. The resulting tokens capture both the semantic meaning and the fine-grained details needed for visual reconstruction, effectively translating the visual content into a language comprehensible to the LLM, and empowering it to perform a wide array of multimodal tasks. Our approach is validated through in-context learning experiments with frozen PaLM 2 and GPT 3.5 on a diverse set of image understanding and generation tasks. Our method marks the first successful attempt to enable a frozen LLM to generate image content while surpassing state-of-the-art performance in image understanding tasks, under the same setting, by over 25%.

José Lezama, Huiwen Chang, Lu Jiang et al. · deepmind

Non-autoregressive generative transformers recently demonstrated impressive image generation performance, and orders of magnitude faster sampling than their autoregressive counterparts. However, optimal parallel sampling from the true joint distribution of visual tokens remains an open challenge. In this paper we introduce Token-Critic, an auxiliary model to guide the sampling of a non-autoregressive generative transformer. Given a masked-and-reconstructed real image, the Token-Critic model is trained to distinguish which visual tokens belong to the original image and which were sampled by the generative transformer. During non-autoregressive iterative sampling, Token-Critic is used to select which tokens to accept and which to reject and resample. Coupled with Token-Critic, a state-of-the-art generative transformer significantly improves its performance, and outperforms recent diffusion models and GANs in terms of the trade-off between generated image quality and diversity, in the challenging class-conditional ImageNet generation.

Tianpeng Bu, Xin Liu, Qihua Chen et al.

While GUI agents have advanced rapidly, they often lack the robustness to recover from their own errors, hindering real-world deployment. To bridge this gap at both the evaluation and data levels, we introduce GUI-RobustEval and propose Robustness-driven Trajectory Synthesis. GUI-RobustEval contains $1,216$ executable test cases that systematically measure error recovery capabilities across a broad and realistic spectrum of error modes. At the data level, RoTS is a scalable synthesis framework that creates $800k$ high-quality data via a tree-based pipeline that proactively discovers diverse error modes and synthesizes corresponding recovery steps. Our two models, RoTS-7B and RoTS-32B, fine-tuned on our dataset, both demonstrate significant gains on GUI-RobustEval and traditional GUI benchmarks. Notably, RoTS-32B achieves state-of-the-art performance on OSWorld, with a $47.4\%$ success rate and a $33.8\%$ All-Pass@4 score, suggesting that improved long-horizon error recovery ability contributes to both robustness and overall performance. Our code is available at https://github.com/AlibabaResearch/RoTS.

Kihyuk Sohn, Albert Shaw, Yuan Hao et al. · deepmind, gatech

We study domain-adaptive image synthesis, the problem of teaching pretrained image generative models a new style or concept from as few as one image to synthesize novel images, to better understand the compositional image synthesis. We present a framework that leverages a pretrained class-conditional generation model and visual prompt tuning. Specifically, we propose a novel source class distilled visual prompt that learns disentangled prompts of semantic (e.g., class) and domain (e.g., style) from a few images. Learned domain prompt is then used to synthesize images of any classes in the style of target domain. We conduct studies on various target domains with the number of images ranging from one to a few to many, and show qualitative results which show the compositional generalization of our method. Moreover, we show that our method can help improve zero-shot domain adaptation classification accuracy.

Yanzhe Zhang, Lu Jiang, Greg Turk et al. · gatech

Text-to-image models, which can generate high-quality images based on textual input, have recently enabled various content-creation tools. Despite significantly affecting a wide range of downstream applications, the distributions of these generated images are still not fully understood, especially when it comes to the potential stereotypical attributes of different genders. In this work, we propose a paradigm (Gender Presentation Differences) that utilizes fine-grained self-presentation attributes to study how gender is presented differently in text-to-image models. By probing gender indicators in the input text (e.g., "a woman" or "a man"), we quantify the frequency differences of presentation-centric attributes (e.g., "a shirt" and "a dress") through human annotation and introduce a novel metric: GEP. Furthermore, we propose an automatic method to estimate such differences. The automatic GEP metric based on our approach yields a higher correlation with human annotations than that based on existing CLIP scores, consistently across three state-of-the-art text-to-image models. Finally, we demonstrate the generalization ability of our metrics in the context of gender stereotypes related to occupations.

Yuanze Lin, Yi-Wen Chen, Yi-Hsuan Tsai et al.

Language has emerged as a natural interface for image editing. In this paper, we introduce a method for region-based image editing driven by textual prompts, without the need for user-provided masks or sketches. Specifically, our approach leverages an existing pre-trained text-to-image model and introduces a bounding box generator to identify the editing regions that are aligned with the textual prompts. We show that this simple approach enables flexible editing that is compatible with current image generation models, and is able to handle complex prompts featuring multiple objects, complex sentences, or lengthy paragraphs. We conduct an extensive user study to compare our method against state-of-the-art methods. The experiments demonstrate the competitive performance of our method in manipulating images with high fidelity and realism that correspond to the provided language descriptions. Our project webpage can be found at: https://yuanze-lin.me/LearnableRegions_page.

Pengzhou Cai, Lu Jiang, Yanxin Li et al.

In this paper, we propose a method, named BRAU-Net, to solve the pubic symphysis-fetal head segmentation task. The method adopts a U-Net-like pure Transformer architecture with bi-level routing attention and skip connections, which effectively learns local-global semantic information. The proposed BRAU-Net was evaluated on transperineal Ultrasound images dataset from the pubic symphysis-fetal head segmentation and angle of progression (FH-PS-AOP) challenge. The results demonstrate that the proposed BRAU-Net achieves comparable a final score. The codes will be available at https://github.com/Caipengzhou/BRAU-Net.

Jieyun Bai, Zihao Zhou, Zhanhong Ou et al.

Segmentation of the fetal and maternal structures, particularly intrapartum ultrasound imaging as advocated by the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) for monitoring labor progression, is a crucial first step for quantitative diagnosis and clinical decision-making. This requires specialized analysis by obstetrics professionals, in a task that i) is highly time- and cost-consuming and ii) often yields inconsistent results. The utility of automatic segmentation algorithms for biometry has been proven, though existing results remain suboptimal. To push forward advancements in this area, the Grand Challenge on Pubic Symphysis-Fetal Head Segmentation (PSFHS) was held alongside the 26th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2023). This challenge aimed to enhance the development of automatic segmentation algorithms at an international scale, providing the largest dataset to date with 5,101 intrapartum ultrasound images collected from two ultrasound machines across three hospitals from two institutions. The scientific community's enthusiastic participation led to the selection of the top 8 out of 179 entries from 193 registrants in the initial phase to proceed to the competition's second stage. These algorithms have elevated the state-of-the-art in automatic PSFHS from intrapartum ultrasound images. A thorough analysis of the results pinpointed ongoing challenges in the field and outlined recommendations for future work. The top solutions and the complete dataset remain publicly available, fostering further advancements in automatic segmentation and biometry for intrapartum ultrasound imaging.

Yanan Xiao, Minyu Liu, Zichen Zhang et al.

Traffic flow prediction is an important part of smart transportation. The goal is to predict future traffic conditions based on historical data recorded by sensors and the traffic network. As the city continues to build, parts of the transportation network will be added or modified. How to accurately predict expanding and evolving long-term streaming networks is of great significance. To this end, we propose a new simulation-based criterion that considers teaching autonomous agents to mimic sensor patterns, planning their next visit based on the sensor's profile (e.g., traffic, speed, occupancy). The data recorded by the sensor is most accurate when the agent can perfectly simulate the sensor's activity pattern. We propose to formulate the problem as a continuous reinforcement learning task, where the agent is the next flow value predictor, the action is the next time-series flow value in the sensor, and the environment state is a dynamically fused representation of the sensor and transportation network. Actions taken by the agent change the environment, which in turn forces the agent's mode to update, while the agent further explores changes in the dynamic traffic network, which helps the agent predict its next visit more accurately. Therefore, we develop a strategy in which sensors and traffic networks update each other and incorporate temporal context to quantify state representations evolving over time.

Hao Wang, Jiqing Zhang, Xin Yang et al.

Camouflaged Object Detection (COD) aims to segment objects that blend seamlessly into complex backgrounds, with growing interest in exploiting additional visual modalities to enhance robustness through complementary information. However, most existing approaches generally rely on modality-specific architectures or customized fusion strategies, which limit scalability and cross-modal generalization. To address this, we propose a novel framework that generates modality-agnostic multi-modal prompts for the Segment Anything Model (SAM), enabling parameter-efficient adaptation to arbitrary auxiliary modalities and significantly improving overall performance on COD tasks. Specifically, we model multi-modal learning through interactions between a data-driven content domain and a knowledge-driven prompt domain, distilling task-relevant cues into unified prompts for SAM decoding. We further introduce a lightweight Mask Refine Module to calibrate coarse predictions by incorporating fine-grained prompt cues, leading to more accurate camouflaged object boundaries. Extensive experiments on RGB-Depth, RGB-Thermal, and RGB-Polarization benchmarks validate the effectiveness and generalization of our modality-agnostic framework.

Lu Jiang, Yuanhan Li, Na Luo et al.

With the development of technology and sharing economy, Airbnb as a famous short-term rental platform, has become the first choice for many young people to select. The issue of Airbnb's pricing has always been a problem worth studying. While the previous studies achieve promising results, there are exists deficiencies to solve. Such as, (1) the feature attributes of rental are not rich enough; (2) the research on rental text information is not deep enough; (3) there are few studies on predicting the rental price combined with the point of interest(POI) around the house. To address the above challenges, we proposes a multi-source information embedding(MSIE) model to predict the rental price of Airbnb. Specifically, we first selects the statistical feature to embed the original rental data. Secondly, we generates the word feature vector and emotional score combination of three different text information to form the text feature embedding. Thirdly, we uses the points of interest(POI) around the rental house information generates a variety of spatial network graphs, and learns the embedding of the network to obtain the spatial feature embedding. Finally, this paper combines the three modules into multi source rental representations, and uses the constructed fully connected neural network to predict the price. The analysis of the experimental results shows the effectiveness of our proposed model.

Lu Jiang, Yibin Wang, Jianan Wang et al.

In this paper, we study the problem of MOOC quality evaluation which is essential for improving the course materials, promoting students' learning efficiency, and benefiting user services. While achieving promising performances, current works still suffer from the complicated interactions and relationships of entities in MOOC platforms. To tackle the challenges, we formulate the problem as a course representation learning task-based and develop an Information-aware Graph Representation Learning(IaGRL) for multi-view MOOC quality evaluation. Specifically, We first build a MOOC Heterogeneous Network (HIN) to represent the interactions and relationships among entities in MOOC platforms. And then we decompose the MOOC HIN into multiple single-relation graphs based on meta-paths to depict the multi-view semantics of courses. The course representation learning can be further converted to a multi-view graph representation task. Different from traditional graph representation learning, the learned course representations are expected to match the following three types of validity: (1) the agreement on expressiveness between the raw course portfolio and the learned course representations; (2) the consistency between the representations in each view and the unified representations; (3) the alignment between the course and MOOC platform representations. Therefore, we propose to exploit mutual information for preserving the validity of course representations. We conduct extensive experiments over real-world MOOC datasets to demonstrate the effectiveness of our proposed method.

Dan Kondratyuk, Lijun Yu, Xiuye Gu et al. · cmu, deepmind

We present VideoPoet, a language model capable of synthesizing high-quality video, with matching audio, from a large variety of conditioning signals. VideoPoet employs a decoder-only transformer architecture that processes multimodal inputs -- including images, videos, text, and audio. The training protocol follows that of Large Language Models (LLMs), consisting of two stages: pretraining and task-specific adaptation. During pretraining, VideoPoet incorporates a mixture of multimodal generative objectives within an autoregressive Transformer framework. The pretrained LLM serves as a foundation that can be adapted for a range of video generation tasks. We present empirical results demonstrating the model's state-of-the-art capabilities in zero-shot video generation, specifically highlighting VideoPoet's ability to generate high-fidelity motions. Project page: http://sites.research.google/videopoet/

Pengzhou Cai, Lu Jiang, Yanxin Li et al.

Pubic symphysis-fetal head segmentation in transperineal ultrasound images plays a critical role for the assessment of fetal head descent and progression. Existing transformer segmentation methods based on sparse attention mechanism use handcrafted static patterns, which leads to great differences in terms of segmentation performance on specific datasets. To address this issue, we introduce a dynamic, query-aware sparse attention mechanism for ultrasound image segmentation. Specifically, we propose a novel method, named BRAU-Net to solve the pubic symphysis-fetal head segmentation task in this paper. The method adopts a U-Net-like encoder-decoder architecture with bi-level routing attention and skip connections, which effectively learns local-global semantic information. In addition, we propose an inverted bottleneck patch expanding (IBPE) module to reduce information loss while performing up-sampling operations. The proposed BRAU-Net is evaluated on FH-PS-AoP and HC18 datasets. The results demonstrate that our method could achieve excellent segmentation results. The code is available on GitHub.

Agrim Gupta, Lijun Yu, Kihyuk Sohn et al. · cmu, deepmind

We present W.A.L.T, a transformer-based approach for photorealistic video generation via diffusion modeling. Our approach has two key design decisions. First, we use a causal encoder to jointly compress images and videos within a unified latent space, enabling training and generation across modalities. Second, for memory and training efficiency, we use a window attention architecture tailored for joint spatial and spatiotemporal generative modeling. Taken together these design decisions enable us to achieve state-of-the-art performance on established video (UCF-101 and Kinetics-600) and image (ImageNet) generation benchmarks without using classifier free guidance. Finally, we also train a cascade of three models for the task of text-to-video generation consisting of a base latent video diffusion model, and two video super-resolution diffusion models to generate videos of $512 \times 896$ resolution at $8$ frames per second.

Curtis G. Northcutt, Lu Jiang, Isaac L. Chuang

Learning exists in the context of data, yet notions of confidence typically focus on model predictions, not label quality. Confident learning (CL) is an alternative approach which focuses instead on label quality by characterizing and identifying label errors in datasets, based on the principles of pruning noisy data, counting with probabilistic thresholds to estimate noise, and ranking examples to train with confidence. Whereas numerous studies have developed these principles independently, here, we combine them, building on the assumption of a class-conditional noise process to directly estimate the joint distribution between noisy (given) labels and uncorrupted (unknown) labels. This results in a generalized CL which is provably consistent and experimentally performant. We present sufficient conditions where CL exactly finds label errors, and show CL performance exceeding seven recent competitive approaches for learning with noisy labels on the CIFAR dataset. Uniquely, the CL framework is not coupled to a specific data modality or model (e.g., we use CL to find several label errors in the presumed error-free MNIST dataset and improve sentiment classification on text data in Amazon Reviews). We also employ CL on ImageNet to quantify ontological class overlap (e.g., estimating 645 "missile" images are mislabeled as their parent class "projectile"), and moderately increase model accuracy (e.g., for ResNet) by cleaning data prior to training. These results are replicable using the open-source cleanlab release.

Lu Jiang, Zhengyuan Zhou, Thomas Leung et al.

Recent deep networks are capable of memorizing the entire data even when the labels are completely random. To overcome the overfitting on corrupted labels, we propose a novel technique of learning another neural network, called MentorNet, to supervise the training of the base deep networks, namely, StudentNet. During training, MentorNet provides a curriculum (sample weighting scheme) for StudentNet to focus on the sample the label of which is probably correct. Unlike the existing curriculum that is usually predefined by human experts, MentorNet learns a data-driven curriculum dynamically with StudentNet. Experimental results demonstrate that our approach can significantly improve the generalization performance of deep networks trained on corrupted training data. Notably, to the best of our knowledge, we achieve the best-published result on WebVision, a large benchmark containing 2.2 million images of real-world noisy labels. The code are at https://github.com/google/mentornet

Po-Yao Huang, Ye Yuan, Zhenzhong Lan et al.

We report on CMU Informedia Lab's system used in Google's YouTube 8 Million Video Understanding Challenge. In this multi-label video classification task, our pipeline achieved 84.675% and 84.662% GAP on our evaluation split and the official test set. We attribute the good performance to three components: 1) Refined video representation learning with residual links and hypercolumns 2) Latent concept mining which captures interactions among concepts. 3) Learning with temporal segments and weighted multi-model ensemble. We conduct experiments to validate and analyze the contribution of our models. We also share some unsuccessful trials leveraging conventional approaches such as recurrent neural networks for video representation learning for this large-scale video dataset. All the codes to reproduce our results are publicly available at https://github.com/Martini09/informedia-yt8m-release.

Yanan Xiao, Yixiang Tang, Zechen Feng et al.

While experience replay is essential for data efficiency in reinforcement learning (RL), standard methods treat the replay buffer as a passive memory system, prioritizing samples based on numerical prediction errors rather than their semantic significance. This approach stands in contrast to human learning, which accelerates mastery by actively abstracting fragmented experiences into behavioral rules. To bridge this gap, we propose Neuro-Symbolic Experience Replay (NSER), a framework that transforms experience replay from a passive sample reuse mechanism into an active engine for knowledge construction. Specifically, NSER addresses the incompatibility between linguistic reasoning and numerical optimization through a novel neuro-symbolic grounding pipeline. It leverages Large Language Models (LLMs) in a zero-shot manner to induce candidate behavioral rules from accumulated trajectories, grounds these insights into differentiable first-order logic representations, and utilizes the resulting symbolic structures to dynamically reweight the replay distribution. By allowing abstract knowledge to directly shape policy optimization, NSER achieves consistent superior sample efficiency and convergence speed across reactive, rule-based, and procedural benchmarks.

Team Seawead, Ceyuan Yang, Zhijie Lin et al.

This technical report presents a cost-efficient strategy for training a video generation foundation model. We present a mid-sized research model with approximately 7 billion parameters (7B) called Seaweed-7B trained from scratch using 665,000 H100 GPU hours. Despite being trained with moderate computational resources, Seaweed-7B demonstrates highly competitive performance compared to contemporary video generation models of much larger size. Design choices are especially crucial in a resource-constrained setting. This technical report highlights the key design decisions that enhance the performance of the medium-sized diffusion model. Empirically, we make two observations: (1) Seaweed-7B achieves performance comparable to, or even surpasses, larger models trained on substantially greater GPU resources, and (2) our model, which exhibits strong generalization ability, can be effectively adapted across a wide range of downstream applications either by lightweight fine-tuning or continue training. See the project page at https://seaweed.video/

Shanchuan Lin, Xin Xia, Yuxi Ren et al.

The diffusion models are widely used for image and video generation, but their iterative generation process is slow and expansive. While existing distillation approaches have demonstrated the potential for one-step generation in the image domain, they still suffer from significant quality degradation. In this work, we propose Adversarial Post-Training (APT) against real data following diffusion pre-training for one-step video generation. To improve the training stability and quality, we introduce several improvements to the model architecture and training procedures, along with an approximated R1 regularization objective. Empirically, our experiments show that our adversarial post-trained model, Seaweed-APT, can generate 2-second, 1280x720, 24fps videos in real time using a single forward evaluation step. Additionally, our model is capable of generating 1024px images in a single step, achieving quality comparable to state-of-the-art methods.

Yu Gao, Haoyuan Guo, Tuyen Hoang et al.

Notable breakthroughs in diffusion modeling have propelled rapid improvements in video generation, yet current foundational model still face critical challenges in simultaneously balancing prompt following, motion plausibility, and visual quality. In this report, we introduce Seedance 1.0, a high-performance and inference-efficient video foundation generation model that integrates several core technical improvements: (i) multi-source data curation augmented with precision and meaningful video captioning, enabling comprehensive learning across diverse scenarios; (ii) an efficient architecture design with proposed training paradigm, which allows for natively supporting multi-shot generation and jointly learning of both text-to-video and image-to-video tasks. (iii) carefully-optimized post-training approaches leveraging fine-grained supervised fine-tuning, and video-specific RLHF with multi-dimensional reward mechanisms for comprehensive performance improvements; (iv) excellent model acceleration achieving ~10x inference speedup through multi-stage distillation strategies and system-level optimizations. Seedance 1.0 can generate a 5-second video at 1080p resolution only with 41.4 seconds (NVIDIA-L20). Compared to state-of-the-art video generation models, Seedance 1.0 stands out with high-quality and fast video generation having superior spatiotemporal fluidity with structural stability, precise instruction adherence in complex multi-subject contexts, native multi-shot narrative coherence with consistent subject representation.

Yuwei Guo, Ceyuan Yang, Ziyan Yang et al.

Recent advances in video generation can produce realistic, minute-long single-shot videos with scalable diffusion transformers. However, real-world narrative videos require multi-shot scenes with visual and dynamic consistency across shots. In this work, we introduce Long Context Tuning (LCT), a training paradigm that expands the context window of pre-trained single-shot video diffusion models to learn scene-level consistency directly from data. Our method expands full attention mechanisms from individual shots to encompass all shots within a scene, incorporating interleaved 3D position embedding and an asynchronous noise strategy, enabling both joint and auto-regressive shot generation without additional parameters. Models with bidirectional attention after LCT can further be fine-tuned with context-causal attention, facilitating auto-regressive generation with efficient KV-cache. Experiments demonstrate single-shot models after LCT can produce coherent multi-shot scenes and exhibit emerging capabilities, including compositional generation and interactive shot extension, paving the way for more practical visual content creation. See https://guoyww.github.io/projects/long-context-video/ for more details.

Hao He, Ceyuan Yang, Shanchuan Lin et al.

This paper introduces CameraCtrl II, a framework that enables large-scale dynamic scene exploration through a camera-controlled video diffusion model. Previous camera-conditioned video generative models suffer from diminished video dynamics and limited range of viewpoints when generating videos with large camera movement. We take an approach that progressively expands the generation of dynamic scenes -- first enhancing dynamic content within individual video clip, then extending this capability to create seamless explorations across broad viewpoint ranges. Specifically, we construct a dataset featuring a large degree of dynamics with camera parameter annotations for training while designing a lightweight camera injection module and training scheme to preserve dynamics of the pretrained models. Building on these improved single-clip techniques, we enable extended scene exploration by allowing users to iteratively specify camera trajectories for generating coherent video sequences. Experiments across diverse scenarios demonstrate that CameraCtrl Ii enables camera-controlled dynamic scene synthesis with substantially wider spatial exploration than previous approaches.

Libin Lan, Pengzhou Cai, Lu Jiang et al.

Accurate medical image segmentation is essential for clinical quantification, disease diagnosis, treatment planning and many other applications. Both convolution-based and transformer-based u-shaped architectures have made significant success in various medical image segmentation tasks. The former can efficiently learn local information of images while requiring much more image-specific inductive biases inherent to convolution operation. The latter can effectively capture long-range dependency at different feature scales using self-attention, whereas it typically encounters the challenges of quadratic compute and memory requirements with sequence length increasing. To address this problem, through integrating the merits of these two paradigms in a well-designed u-shaped architecture, we propose a hybrid yet effective CNN-Transformer network, named BRAU-Net++, for an accurate medical image segmentation task. Specifically, BRAU-Net++ uses bi-level routing attention as the core building block to design our u-shaped encoder-decoder structure, in which both encoder and decoder are hierarchically constructed, so as to learn global semantic information while reducing computational complexity. Furthermore, this network restructures skip connection by incorporating channel-spatial attention which adopts convolution operations, aiming to minimize local spatial information loss and amplify global dimension-interaction of multi-scale features. Extensive experiments on three public benchmark datasets demonstrate that our proposed approach surpasses other state-of-the-art methods including its baseline: BRAU-Net under almost all evaluation metrics. We achieve the average Dice-Similarity Coefficient (DSC) of 82.47, 90.10, and 92.94 on Synapse multi-organ segmentation, ISIC-2018 Challenge, and CVC-ClinicDB, as well as the mIoU of 84.01 and 88.17 on ISIC-2018 Challenge and CVC-ClinicDB, respectively.

Jianyi Wang, Zhijie Lin, Meng Wei et al.

Video restoration poses non-trivial challenges in maintaining fidelity while recovering temporally consistent details from unknown degradations in the wild. Despite recent advances in diffusion-based restoration, these methods often face limitations in generation capability and sampling efficiency. In this work, we present SeedVR, a diffusion transformer designed to handle real-world video restoration with arbitrary length and resolution. The core design of SeedVR lies in the shifted window attention that facilitates effective restoration on long video sequences. SeedVR further supports variable-sized windows near the boundary of both spatial and temporal dimensions, overcoming the resolution constraints of traditional window attention. Equipped with contemporary practices, including causal video autoencoder, mixed image and video training, and progressive training, SeedVR achieves highly-competitive performance on both synthetic and real-world benchmarks, as well as AI-generated videos. Extensive experiments demonstrate SeedVR's superiority over existing methods for generic video restoration.

Xinhao Zhang, Zaitian Wang, Lu Jiang et al.

In this paper, we propose a novel feature weighting method to address the limitation of existing feature processing methods for tabular data. Typically the existing methods assume equal importance across all samples and features in one dataset. This simplified processing methods overlook the unique contributions of each feature, and thus may miss important feature information. As a result, it leads to suboptimal performance in complex datasets with rich features. To address this problem, we introduce Tabular Feature Weighting with Transformer, a novel feature weighting approach for tabular data. Our method adopts Transformer to capture complex feature dependencies and contextually assign appropriate weights to discrete and continuous features. Besides, we employ a reinforcement learning strategy to further fine-tune the weighting process. Our extensive experimental results across various real-world datasets and diverse downstream tasks show the effectiveness of TFWT and highlight the potential for enhancing feature weighting in tabular data analysis.

Hsin-Ping Huang, Yu-Chuan Su, Deqing Sun et al. · deepmind

Text-to-video generation has shown promising results. However, by taking only natural languages as input, users often face difficulties in providing detailed information to precisely control the model's output. In this work, we propose fine-grained controllable video generation (FACTOR) to achieve detailed control. Specifically, FACTOR aims to control objects' appearances and context, including their location and category, in conjunction with the text prompt. To achieve detailed control, we propose a unified framework to jointly inject control signals into the existing text-to-video model. Our model consists of a joint encoder and adaptive cross-attention layers. By optimizing the encoder and the inserted layer, we adapt the model to generate videos that are aligned with both text prompts and fine-grained control. Compared to existing methods relying on dense control signals such as edge maps, we provide a more intuitive and user-friendly interface to allow object-level fine-grained control. Our method achieves controllability of object appearances without finetuning, which reduces the per-subject optimization efforts for the users. Extensive experiments on standard benchmark datasets and user-provided inputs validate that our model obtains a 70% improvement in controllability metrics over competitive baselines.

Shanchuan Lin, Ceyuan Yang, Hao He et al.

Existing large-scale video generation models are computationally intensive, preventing adoption in real-time and interactive applications. In this work, we propose autoregressive adversarial post-training (AAPT) to transform a pre-trained latent video diffusion model into a real-time, interactive video generator. Our model autoregressively generates a latent frame at a time using a single neural function evaluation (1NFE). The model can stream the result to the user in real time and receive interactive responses as controls to generate the next latent frame. Unlike existing approaches, our method explores adversarial training as an effective paradigm for autoregressive generation. This not only allows us to design an architecture that is more efficient for one-step generation while fully utilizing the KV cache, but also enables training the model in a student-forcing manner that proves to be effective in reducing error accumulation during long video generation. Our experiments demonstrate that our 8B model achieves real-time, 24fps, streaming video generation at 736x416 resolution on a single H100, or 1280x720 on 8xH100 up to a minute long (1440 frames). Visit our research website at https://seaweed-apt.com/2

Gwanghyun Kim, Alonso Martinez, Yu-Chuan Su et al. · cmu

Training diffusion models for audiovisual sequences allows for a range of generation tasks by learning conditional distributions of various input-output combinations of the two modalities. Nevertheless, this strategy often requires training a separate model for each task which is expensive. Here, we propose a novel training approach to effectively learn arbitrary conditional distributions in the audiovisual space.Our key contribution lies in how we parameterize the diffusion timestep in the forward diffusion process. Instead of the standard fixed diffusion timestep, we propose applying variable diffusion timesteps across the temporal dimension and across modalities of the inputs. This formulation offers flexibility to introduce variable noise levels for various portions of the input, hence the term mixture of noise levels. We propose a transformer-based audiovisual latent diffusion model and show that it can be trained in a task-agnostic fashion using our approach to enable a variety of audiovisual generation tasks at inference time. Experiments demonstrate the versatility of our method in tackling cross-modal and multimodal interpolation tasks in the audiovisual space. Notably, our proposed approach surpasses baselines in generating temporally and perceptually consistent samples conditioned on the input. Project page: avdit2024.github.io

Jiaming Han, Hao Chen, Yang Zhao et al.

This paper presents a multimodal framework that attempts to unify visual understanding and generation within a shared discrete semantic representation. At its core is the Text-Aligned Tokenizer (TA-Tok), which converts images into discrete tokens using a text-aligned codebook projected from a large language model's (LLM) vocabulary. By integrating vision and text into a unified space with an expanded vocabulary, our multimodal LLM, Tar, enables cross-modal input and output through a shared interface, without the need for modality-specific designs. Additionally, we propose scale-adaptive encoding and decoding to balance efficiency and visual detail, along with a generative de-tokenizer to produce high-fidelity visual outputs. To address diverse decoding needs, we utilize two complementary de-tokenizers: a fast autoregressive model and a diffusion-based model. To enhance modality fusion, we investigate advanced pre-training tasks, demonstrating improvements in both visual understanding and generation. Experiments across benchmarks show that Tar matches or surpasses existing multimodal LLM methods, achieving faster convergence and greater training efficiency. Code, models, and data are available at https://tar.csuhan.com

Zhaofan Zhang, Yanan Xiao, Lu Jiang et al.

In the realm of human mobility, the decision-making process for selecting the next-visit location is intricately influenced by a trade-off between spatial and temporal constraints, which are reflective of individual needs and preferences. This trade-off, however, varies across individuals, making the modeling of these spatial-temporal dynamics a formidable challenge. To address the problem, in this work, we introduce the "Spatial-temporal Induced Hierarchical Reinforcement Learning" (STI-HRL) framework, for capturing the interplay between spatial and temporal factors in human mobility decision-making. Specifically, STI-HRL employs a two-tiered decision-making process: the low-level focuses on disentangling spatial and temporal preferences using dedicated agents, while the high-level integrates these considerations to finalize the decision. To complement the hierarchical decision setting, we construct a hypergraph to organize historical data, encapsulating the multi-aspect semantics of human mobility. We propose a cross-channel hypergraph embedding module to learn the representations as the states to facilitate the decision-making cycle. Our extensive experiments on two real-world datasets validate the superiority of STI-HRL over state-of-the-art methods in predicting users' next visits across various performance metrics.

Shengqu Cai, Ceyuan Yang, Lvmin Zhang et al. · stanford

Long video generation is fundamentally a long context memory problem: models must retain and retrieve salient events across a long range without collapsing or drifting. However, scaling diffusion transformers to generate long-context videos is fundamentally limited by the quadratic cost of self-attention, which makes memory and computation intractable and difficult to optimize for long sequences. We recast long-context video generation as an internal information retrieval task and propose a simple, learnable sparse attention routing module, Mixture of Contexts (MoC), as an effective long-term memory retrieval engine. In MoC, each query dynamically selects a few informative chunks plus mandatory anchors (caption, local windows) to attend to, with causal routing that prevents loop closures. As we scale the data and gradually sparsify the routing, the model allocates compute to salient history, preserving identities, actions, and scenes over minutes of content. Efficiency follows as a byproduct of retrieval (near-linear scaling), which enables practical training and synthesis, and the emergence of memory and consistency at the scale of minutes.

Junfei Xiao, Ceyuan Yang, Lvmin Zhang et al. · stanford

We present Captain Cinema, a generation framework for short movie generation. Given a detailed textual description of a movie storyline, our approach firstly generates a sequence of keyframes that outline the entire narrative, which ensures long-range coherence in both the storyline and visual appearance (e.g., scenes and characters). We refer to this step as top-down keyframe planning. These keyframes then serve as conditioning signals for a video synthesis model, which supports long context learning, to produce the spatio-temporal dynamics between them. This step is referred to as bottom-up video synthesis. To support stable and efficient generation of multi-scene long narrative cinematic works, we introduce an interleaved training strategy for Multimodal Diffusion Transformers (MM-DiT), specifically adapted for long-context video data. Our model is trained on a specially curated cinematic dataset consisting of interleaved data pairs. Our experiments demonstrate that Captain Cinema performs favorably in the automated creation of visually coherent and narrative consistent short movies in high quality and efficiency. Project page: https://thecinema.ai

Jianyi Wang, Shanchuan Lin, Zhijie Lin et al.

Recent advances in diffusion-based video restoration (VR) demonstrate significant improvement in visual quality, yet yield a prohibitive computational cost during inference. While several distillation-based approaches have exhibited the potential of one-step image restoration, extending existing approaches to VR remains challenging and underexplored, particularly when dealing with high-resolution video in real-world settings. In this work, we propose a one-step diffusion-based VR model, termed as SeedVR2, which performs adversarial VR training against real data. To handle the challenging high-resolution VR within a single step, we introduce several enhancements to both model architecture and training procedures. Specifically, an adaptive window attention mechanism is proposed, where the window size is dynamically adjusted to fit the output resolutions, avoiding window inconsistency observed under high-resolution VR using window attention with a predefined window size. To stabilize and improve the adversarial post-training towards VR, we further verify the effectiveness of a series of losses, including a proposed feature matching loss without significantly sacrificing training efficiency. Extensive experiments show that SeedVR2 can achieve comparable or even better performance compared with existing VR approaches in a single step.

Qi Zhao, Xingyu Ni, Ziyu Wang et al.

We investigate how to enhance the physical fidelity of video generation models by leveraging synthetic videos derived from computer graphics pipelines. These rendered videos respect real-world physics, such as maintaining 3D consistency, and serve as a valuable resource that can potentially improve video generation models. To harness this potential, we propose a solution that curates and integrates synthetic data while introducing a method to transfer its physical realism to the model, significantly reducing unwanted artifacts. Through experiments on three representative tasks emphasizing physical consistency, we demonstrate its efficacy in enhancing physical fidelity. While our model still lacks a deep understanding of physics, our work offers one of the first empirical demonstrations that synthetic video enhances physical fidelity in video synthesis. Website: https://kevinz8866.github.io/simulation/

Junfei Xiao, Feng Cheng, Lu Qi et al.

Recent video generation models have shown promising results in producing high-quality video clips lasting several seconds. However, these models face challenges in generating long sequences that convey clear and informative events, limiting their ability to support coherent narrations. In this paper, we present a large-scale cooking video dataset designed to advance long-form narrative generation in the cooking domain. We validate the quality of our proposed dataset in terms of visual fidelity and textual caption accuracy using state-of-the-art Vision-Language Models (VLMs) and video generation models, respectively. We further introduce a Long Narrative Video Director to enhance both visual and semantic coherence in generated videos and emphasize the role of aligning visual embeddings to achieve improved overall video quality. Our method demonstrates substantial improvements in generating visually detailed and semantically aligned keyframes, supported by finetuning techniques that integrate text and image embeddings within the video generation process. Project page: https://videoauteur.github.io/

Libin Lan, Lu Jiang, Tianshu Yu et al.

Video summarization aims to generate a compact, informative, and representative synopsis of raw videos, which is crucial for browsing, analyzing, and understanding video content. Dominant approaches in video summarization primarily rely on recurrent or convolutional neural networks, and more recently on encoder-only transformer architectures. However, these methods typically suffer from several limitations in parallelism, modeling long-range dependencies, and providing explicit generative capabilities. To address these issues, we propose a transformer-like architecture named FullTransNet with two-fold ideas. First, it uses a full transformer with an encoder-decoder structure as an alternative architecture for video summarization. As the full transformer is specifically designed for sequence transduction tasks, its direct application to video summarization is both intuitive and effective. Second, it replaces the standard full attention mechanism with a combination of local and global sparse attention, enabling the model to capture long-range dependencies while significantly reducing computational costs. This local-global sparse attention is applied exclusively at the encoder side, where the majority of computations occur, further enhancing efficiency. Extensive experiments on two widely used benchmark datasets, SumMe and TVSum, demonstrate that our model achieves F-scores of 54.4% and 63.9%, respectively, while maintaining relatively low computational and memory requirements. These results surpass the second-best performing methods by 0.1% and 0.3%, respectively, verifying the effectiveness and efficiency of FullTransNet.

Leigang Qu, Feng Cheng, Ziyan Yang et al.

In-context image editing aims to modify images based on a contextual sequence comprising text and previously generated images. Existing methods typically depend on task-specific pipelines and expert models (e.g., segmentation and inpainting) to curate training data. In this work, we explore whether an in-context image editing model can be learned directly from videos. We introduce a scalable approach to annotate videos as interleaved multimodal sequences. To effectively learn from this data, we design a block-causal diffusion transformer trained on three proxy tasks: next-image prediction, current segmentation prediction, and next-segmentation prediction. Additionally, we propose a novel multi-turn image editing benchmark to advance research in this area. Extensive experiments demonstrate that our model exhibits strong in-context image editing capabilities and achieves state-of-the-art results on two multi-turn image editing benchmarks. Despite being trained exclusively on videos, our model also shows promising abilities in multi-concept composition, story generation, and chain-of-editing applications.

Yuanxin Liu, Rui Zhu, Shuhuai Ren et al.

With the rapid growth of video generative models (VGMs), it is essential to develop reliable and comprehensive automatic metrics for AI-generated videos (AIGVs). Existing methods either use off-the-shelf models optimized for other tasks or rely on human assessment data to train specialized evaluators. These approaches are constrained to specific evaluation aspects and are difficult to scale with the increasing demands for finer-grained and more comprehensive evaluations. To address this issue, this work investigates the feasibility of using multimodal large language models (MLLMs) as a unified evaluator for AIGVs, leveraging their strong visual perception and language understanding capabilities. To evaluate the performance of automatic metrics in unified AIGV evaluation, we introduce a benchmark called UVE-Bench. UVE-Bench collects videos generated by state-of-the-art VGMs and provides pairwise human preference annotations across 15 evaluation aspects. Using UVE-Bench, we extensively evaluate 18 MLLMs. Our empirical results suggest that while advanced MLLMs (e.g., Qwen2VL-72B and InternVL2.5-78B) still lag behind human evaluators, they demonstrate promising ability in unified AIGV evaluation, significantly surpassing existing specialized evaluation methods. Additionally, we conduct an in-depth analysis of key design choices that impact the performance of MLLM-driven evaluators, offering valuable insights for future research on AIGV evaluation.

Weichen Yu, Ziyan Yang, Shanchuan Lin et al.

In text-to-image (T2I) generation, a prevalent training technique involves utilizing Vision Language Models (VLMs) for image re-captioning. Even though VLMs are known to exhibit hallucination, generating descriptive content that deviates from the visual reality, the ramifications of such caption hallucinations on T2I generation performance remain under-explored. Through our empirical investigation, we first establish a comprehensive dataset comprising VLM-generated captions, and then systematically analyze how caption hallucination influences generation outcomes. Our findings reveal that (1) the disparities in caption quality persistently impact model outputs during fine-tuning. (2) VLMs confidence scores serve as reliable indicators for detecting and characterizing noise-related patterns in the data distribution. (3) even subtle variations in caption fidelity have significant effects on the quality of learned representations. These findings collectively emphasize the profound impact of caption quality on model performance and highlight the need for more sophisticated robust training algorithm in T2I. In response to these observations, we propose a approach leveraging VLM confidence score to mitigate caption noise, thereby enhancing the robustness of T2I models against hallucination in caption.

Rohan Choudhury, Shanchuan Lin, Jianyi Wang et al.

Diffusion-based super-resolution (SR) is a key component in video generation and video restoration, but is slow and expensive, limiting scalability to higher resolutions and longer videos. Our key insight is that many regions in video are inherently low-detail and gain little from refinement, yet current methods process all pixels uniformly. To take advantage of this, we propose SkipSR, a simple framework for accelerating video SR by identifying low-detail regions directly from low-resolution input, then skipping computation on them entirely, only super-resolving the areas that require refinement. This simple yet effective strategy preserves perceptual quality in both standard and one-step diffusion SR models while significantly reducing computation. In standard SR benchmarks, our method achieves up to 60% faster end-to-end latency than prior models on 720p videos with no perceptible loss in quality. Video demos are available at https://rccchoudhury.github.io/skipsr/

Pengjiang Li, Zaitian Wang, Xinhao Zhang et al.

Topic discovery in scientific literature provides valuable insights for researchers to identify emerging trends and explore new avenues for investigation, facilitating easier scientific information retrieval. Many machine learning methods, particularly deep embedding techniques, have been applied to discover research topics. However, most existing topic discovery methods rely on word embedding to capture the semantics and lack a comprehensive understanding of scientific publications, struggling with complex, high-dimensional text relationships. Inspired by the exceptional comprehension of textual information by large language models (LLMs), we propose an advanced topic discovery method enhanced by LLMs to improve scientific topic identification, namely SciTopic. Specifically, we first build a textual encoder to capture the content from scientific publications, including metadata, title, and abstract. Next, we construct a space optimization module that integrates entropy-based sampling and triplet tasks guided by LLMs, enhancing the focus on thematic relevance and contextual intricacies between ambiguous instances. Then, we propose to fine-tune the textual encoder based on the guidance from the LLMs by optimizing the contrastive loss of the triplets, forcing the text encoder to better discriminate instances of different topics. Finally, extensive experiments conducted on three real-world datasets of scientific publications demonstrate that SciTopic outperforms the state-of-the-art (SOTA) scientific topic discovery methods, enabling researchers to gain deeper and faster insights.

Qi Hao, Runchang Liang, Yue Gao et al.

Variable Subset Forecasting (VSF) refers to a unique scenario in multivariate time series forecasting, where available variables in the inference phase are only a subset of the variables in the training phase. VSF presents significant challenges as the entire time series may be missing, and neither inter- nor intra-variable correlations persist. Such conditions impede the effectiveness of traditional imputation methods, primarily focusing on filling in individual missing data points. Inspired by the principle of feature engineering that not all variables contribute positively to forecasting, we propose Task-Oriented Imputation for VSF (TOI-VSF), a novel framework shifts the focus from accurate data recovery to directly support the downstream forecasting task. TOI-VSF incorporates a self-supervised imputation module, agnostic to the forecasting model, designed to fill in missing variables while preserving the vital characteristics and temporal patterns of time series data. Additionally, we implement a joint learning strategy for imputation and forecasting, ensuring that the imputation process is directly aligned with and beneficial to the forecasting objective. Extensive experiments across four datasets demonstrate the superiority of TOI-VSF, outperforming baseline methods by $15\%$ on average.