Yuying Ge

Alex Jinpeng Wang, Yixiao Ge, Rui Yan et al. · tencent-ai

Mainstream Video-Language Pre-training models \cite{actbert,clipbert,violet} consist of three parts, a video encoder, a text encoder, and a video-text fusion Transformer. They pursue better performance via utilizing heavier unimodal encoders or multimodal fusion Transformers, resulting in increased parameters with lower efficiency in downstream tasks. In this work, we for the first time introduce an end-to-end video-language model, namely \textit{all-in-one Transformer}, that embeds raw video and textual signals into joint representations using a unified backbone architecture. We argue that the unique temporal information of video data turns out to be a key barrier hindering the design of a modality-agnostic Transformer. To overcome the challenge, we introduce a novel and effective token rolling operation to encode temporal representations from video clips in a non-parametric manner. The careful design enables the representation learning of both video-text multimodal inputs and unimodal inputs using a unified backbone model. Our pre-trained all-in-one Transformer is transferred to various downstream video-text tasks after fine-tuning, including text-video retrieval, video-question answering, multiple choice and visual commonsense reasoning. State-of-the-art performances with the minimal model FLOPs on nine datasets demonstrate the superiority of our method compared to the competitive counterparts. The code and pretrained model have been released in https://github.com/showlab/all-in-one.

Bohao Li, Yuying Ge, Yixiao Ge et al. · tencent-ai

Multimodal large language models (MLLMs), building upon the foundation of powerful large language models (LLMs), have recently demonstrated exceptional capabilities in generating not only texts but also images given interleaved multimodal inputs (acting like a combination of GPT-4V and DALL-E 3). However, existing MLLM benchmarks remain limited to assessing only models' comprehension ability of single image-text inputs, failing to keep up with the strides made in MLLMs. A comprehensive benchmark is imperative for investigating the progress and uncovering the limitations of current MLLMs. In this work, we categorize the capabilities of MLLMs into hierarchical levels from $L_0$ to $L_4$ based on the modalities they can accept and generate, and propose SEED-Bench-2, a comprehensive benchmark that evaluates the \textbf{hierarchical} capabilities of MLLMs. Specifically, SEED-Bench-2 comprises 24K multiple-choice questions with accurate human annotations, which spans 27 dimensions, including the evaluation of both text and image generation. Multiple-choice questions with groundtruth options derived from human annotation enables an objective and efficient assessment of model performance, eliminating the need for human or GPT intervention during evaluation. We further evaluate the performance of 23 prominent open-source MLLMs and summarize valuable observations. By revealing the limitations of existing MLLMs through extensive evaluations, we aim for SEED-Bench-2 to provide insights that will motivate future research towards the goal of General Artificial Intelligence. Dataset and evaluation code are available at \href{https://github.com/AILab-CVC/SEED-Bench}

Weixian Lei, Yixiao Ge, Kun Yi et al. · tencent-ai

Aiming to advance AI agents, large foundation models significantly improve reasoning and instruction execution, yet the current focus on vision and language neglects the potential of perceiving diverse modalities in open-world environments. However, the success of data-driven vision and language models is costly or even infeasible to be reproduced for rare modalities. In this paper, we present ViT-Lens-2 that facilitates efficient omni-modal representation learning by perceiving novel modalities with a pretrained ViT and aligning them to a pre-defined space. Specifically, the modality-specific lens is tuned to project any-modal signals to an intermediate embedding space, which are then processed by a strong ViT with pre-trained visual knowledge. The encoded representations are optimized toward aligning with the modal-independent space, pre-defined by off-the-shelf foundation models. ViT-Lens-2 provides a unified solution for representation learning of increasing modalities with two appealing advantages: (i) Unlocking the great potential of pretrained ViTs to novel modalities effectively with efficient data regime; (ii) Enabling emergent downstream capabilities through modality alignment and shared ViT parameters. We tailor ViT-Lens-2 to learn representations for 3D point cloud, depth, audio, tactile and EEG, and set new state-of-the-art results across various understanding tasks, such as zero-shot classification. By seamlessly integrating ViT-Lens-2 into Multimodal Foundation Models, we enable Any-modality to Text and Image Generation in a zero-shot manner. Code and models are available at https://github.com/TencentARC/ViT-Lens.

Ziyun Zeng, Yuying Ge, Xihui Liu et al. · tencent-ai

Pre-training on large-scale video data has become a common recipe for learning transferable spatiotemporal representations in recent years. Despite some progress, existing methods are mostly limited to highly curated datasets (e.g., K400) and exhibit unsatisfactory out-of-the-box representations. We argue that it is due to the fact that they only capture pixel-level knowledge rather than spatiotemporal semantics, which hinders further progress in video understanding. Inspired by the great success of image-text pre-training (e.g., CLIP), we take the first step to exploit language semantics to boost transferable spatiotemporal representation learning. We introduce a new pretext task, Turning to Video for Transcript Sorting (TVTS), which sorts shuffled ASR scripts by attending to learned video representations. We do not rely on descriptive captions and learn purely from video, i.e., leveraging the natural transcribed speech knowledge to provide noisy but useful semantics over time. Our method enforces the vision model to contextualize what is happening over time so that it can re-organize the narrative transcripts, and can seamlessly apply to large-scale uncurated video data in the real world. Our method demonstrates strong out-of-the-box spatiotemporal representations on diverse benchmarks, e.g., +13.6% gains over VideoMAE on SSV2 via linear probing. The code is available at https://github.com/TencentARC/TVTS.

Bohao Li, Rui Wang, Guangzhi Wang et al. · tencent-ai

Based on powerful Large Language Models (LLMs), recent generative Multimodal Large Language Models (MLLMs) have gained prominence as a pivotal research area, exhibiting remarkable capability for both comprehension and generation. In this work, we address the evaluation of generative comprehension in MLLMs as a preliminary step towards a comprehensive assessment of generative models, by introducing a benchmark named SEED-Bench. SEED-Bench consists of 19K multiple choice questions with accurate human annotations (x 6 larger than existing benchmarks), which spans 12 evaluation dimensions including the comprehension of both the image and video modality. We develop an advanced pipeline for generating multiple-choice questions that target specific evaluation dimensions, integrating both automatic filtering and manual verification processes. Multiple-choice questions with groundtruth options derived from human annotation enables an objective and efficient assessment of model performance, eliminating the need for human or GPT intervention during evaluation. We further evaluate the performance of 18 models across all 12 dimensions, covering both the spatial and temporal understanding. By revealing the limitations of existing MLLMs through evaluation results, we aim for SEED-Bench to provide insights for motivating future research. We will launch and consistently maintain a leaderboard to provide a platform for the community to assess and investigate model capability.

Keqiang Sun, Junting Pan, Yuying Ge et al. · pku

While recent advancements in vision-language models have had a transformative impact on multi-modal comprehension, the extent to which these models possess the ability to comprehend generated images remains uncertain. Synthetic images, in comparison to real data, encompass a higher level of diversity in terms of both content and style, thereby presenting significant challenges for the models to fully grasp. In light of this challenge, we introduce a comprehensive dataset, referred to as JourneyDB, that caters to the domain of generative images within the context of multi-modal visual understanding. Our meticulously curated dataset comprises 4 million distinct and high-quality generated images, each paired with the corresponding text prompts that were employed in their creation. Furthermore, we additionally introduce an external subset with results of another 22 text-to-image generative models, which makes JourneyDB a comprehensive benchmark for evaluating the comprehension of generated images. On our dataset, we have devised four benchmarks to assess the performance of generated image comprehension in relation to both content and style interpretation. These benchmarks encompass prompt inversion, style retrieval, image captioning, and visual question answering. Lastly, we evaluate the performance of state-of-the-art multi-modal models when applied to the JourneyDB dataset, providing a comprehensive analysis of their strengths and limitations in comprehending generated content. We anticipate that the proposed dataset and benchmarks will facilitate further research in the field of generative content understanding. The dataset is publicly available at https://journeydb.github.io.

Yuying Ge, Sijie Zhao, Ziyun Zeng et al. · tencent-ai

The great success of Large Language Models (LLMs) has expanded the potential of multimodality, contributing to the gradual evolution of General Artificial Intelligence (AGI). A true AGI agent should not only possess the capability to perform predefined multi-tasks but also exhibit emergent abilities in an open-world context. However, despite the considerable advancements made by recent multimodal LLMs, they still fall short in effectively unifying comprehension and generation tasks, let alone open-world emergent abilities. We contend that the key to overcoming the present impasse lies in enabling text and images to be represented and processed interchangeably within a unified autoregressive Transformer. To this end, we introduce SEED, an elaborate image tokenizer that empowers LLMs with the ability to SEE and Draw at the same time. We identify two crucial design principles: (1) Image tokens should be independent of 2D physical patch positions and instead be produced with a 1D causal dependency, exhibiting intrinsic interdependence that aligns with the left-to-right autoregressive prediction mechanism in LLMs. (2) Image tokens should capture high-level semantics consistent with the degree of semantic abstraction in words, and be optimized for both discriminativeness and reconstruction during the tokenizer training phase. With SEED tokens, LLM is able to perform scalable multimodal autoregression under its original training recipe, i.e., next-word prediction. SEED-LLaMA is therefore produced by large-scale pretraining and instruction tuning on the interleaved textual and visual data, demonstrating impressive performance on a broad range of multimodal comprehension and generation tasks. More importantly, SEED-LLaMA has exhibited compositional emergent abilities such as multi-turn in-context multimodal generation, acting like your AI assistant.

Yuying Ge, Yixiao Ge, Ziyun Zeng et al. · tencent-ai

We present SEED, an elaborate image tokenizer that empowers Large Language Models (LLMs) with the emergent ability to SEE and Draw at the same time. Research on image tokenizers has previously reached an impasse, as frameworks employing quantized visual tokens have lost prominence due to subpar performance and convergence in multimodal comprehension (compared to BLIP-2, etc.) or generation (compared to Stable Diffusion, etc.). Despite the limitations, we remain confident in its natural capacity to unify visual and textual representations, facilitating scalable multimodal training with LLM's original recipe. In this study, we identify two crucial principles for the architecture and training of SEED that effectively ease subsequent alignment with LLMs. (1) Image tokens should be independent of 2D physical patch positions and instead be produced with a 1D causal dependency, exhibiting intrinsic interdependence that aligns with the left-to-right autoregressive prediction mechanism in LLMs. (2) Image tokens should capture high-level semantics consistent with the degree of semantic abstraction in words, and be optimized for both discriminativeness and reconstruction during the tokenizer training phase. As a result, the off-the-shelf LLM is able to perform both image-to-text and text-to-image generation by incorporating our SEED through efficient LoRA tuning. Comprehensive multimodal pretraining and instruction tuning, which may yield improved results, are reserved for future investigation. This version of SEED was trained in 5.7 days using only 64 V100 GPUs and 5M publicly available image-text pairs. Our preliminary study emphasizes the great potential of discrete visual tokens in versatile multimodal LLMs and the importance of proper image tokenizers in broader research.

Shuai Yang, Yuying Ge, Yang Li et al. · tencent-ai

With the remarkable advancements in image generation and open-form text generation, the creation of interleaved image-text content has become an increasingly intriguing field. Multimodal story generation, characterized by producing narrative texts and vivid images in an interleaved manner, has emerged as a valuable and practical task with broad applications. However, this task poses significant challenges, as it necessitates the comprehension of the complex interplay between texts and images, and the ability to generate long sequences of coherent, contextually relevant texts and visuals. In this work, we propose SEED-Story, a novel method that leverages a Multimodal Large Language Model (MLLM) to generate extended multimodal stories. Our model, built upon the powerful comprehension capability of MLLM, predicts text tokens as well as visual tokens, which are subsequently processed with an adapted visual de-tokenizer to produce images with consistent characters and styles. We further propose multimodal attention sink mechanism to enable the generation of stories with up to 25 sequences (only 10 for training) in a highly efficient autoregressive manner. Additionally, we present a large-scale and high-resolution dataset named StoryStream for training our model and quantitatively evaluating the task of multimodal story generation in various aspects.

Yuying Ge, Yixiao Ge, Xihui Liu et al. · tencent-ai

Dominant pre-training work for video-text retrieval mainly adopt the "dual-encoder" architectures to enable efficient retrieval, where two separate encoders are used to contrast global video and text representations, but ignore detailed local semantics. The recent success of image BERT pre-training with masked visual modeling that promotes the learning of local visual context, motivates a possible solution to address the above limitation. In this work, we for the first time investigate masked visual modeling in video-text pre-training with the "dual-encoder" architecture. We perform Masked visual modeling with Injected LanguagE Semantics (MILES) by employing an extra snapshot video encoder as an evolving "tokenizer" to produce reconstruction targets for masked video patch prediction. Given the corrupted video, the video encoder is trained to recover text-aligned features of the masked patches via reasoning with the visible regions along the spatial and temporal dimensions, which enhances the discriminativeness of local visual features and the fine-grained cross-modality alignment. Our method outperforms state-of-the-art methods for text-to-video retrieval on four datasets with both zero-shot and fine-tune evaluation protocols. Our approach also surpasses the baseline models significantly on zero-shot action recognition, which can be cast as video-to-text retrieval.

Junting Pan, Ziyi Lin, Yuying Ge et al.

Video Question Answering (VideoQA) has been significantly advanced from the scaling of recent Large Language Models (LLMs). The key idea is to convert the visual information into the language feature space so that the capacity of LLMs can be fully exploited. Existing VideoQA methods typically take two paradigms: (1) learning cross-modal alignment, and (2) using an off-the-shelf captioning model to describe the visual data. However, the first design needs costly training on many extra multi-modal data, whilst the second is further limited by limited domain generalization. To address these limitations, a simple yet effective Retrieving-to-Answer (R2A) framework is proposed.Given an input video, R2A first retrieves a set of semantically similar texts from a generic text corpus using a pre-trained multi-modal model (e.g., CLIP). With both the question and the retrieved texts, a LLM (e.g., DeBERTa) can be directly used to yield a desired answer. Without the need for cross-modal fine-tuning, R2A allows for all the key components (e.g., LLM, retrieval model, and text corpus) to plug-and-play. Extensive experiments on several VideoQA benchmarks show that despite with 1.3B parameters and no fine-tuning, our R2A can outperform the 61 times larger Flamingo-80B model even additionally trained on nearly 2.1B multi-modal data.

Yanjie Ze, Ge Yan, Yueh-Hua Wu et al.

It is a long-standing problem in robotics to develop agents capable of executing diverse manipulation tasks from visual observations in unstructured real-world environments. To achieve this goal, the robot needs to have a comprehensive understanding of the 3D structure and semantics of the scene. In this work, we present $\textbf{GNFactor}$, a visual behavior cloning agent for multi-task robotic manipulation with $\textbf{G}$eneralizable $\textbf{N}$eural feature $\textbf{F}$ields. GNFactor jointly optimizes a generalizable neural field (GNF) as a reconstruction module and a Perceiver Transformer as a decision-making module, leveraging a shared deep 3D voxel representation. To incorporate semantics in 3D, the reconstruction module utilizes a vision-language foundation model ($\textit{e.g.}$, Stable Diffusion) to distill rich semantic information into the deep 3D voxel. We evaluate GNFactor on 3 real robot tasks and perform detailed ablations on 10 RLBench tasks with a limited number of demonstrations. We observe a substantial improvement of GNFactor over current state-of-the-art methods in seen and unseen tasks, demonstrating the strong generalization ability of GNFactor. Our project website is https://yanjieze.com/GNFactor/ .

Jun Zhang, Teng Wang, Yuying Ge et al.

This paper does not introduce a novel method but instead establishes a straightforward, incremental, yet essential baseline for video temporal grounding (VTG), a core capability in video understanding. While multimodal large language models (MLLMs) excel at various video understanding tasks, the recipes for optimizing them for VTG remain under-explored. In this paper, we present TimeLens, a systematic investigation into building MLLMs with strong VTG ability, along two primary dimensions: data quality and algorithmic design. We first expose critical quality issues in existing VTG benchmarks and introduce TimeLens-Bench, comprising meticulously re-annotated versions of three popular benchmarks with strict quality criteria. Our analysis reveals dramatic model re-rankings compared to legacy benchmarks, confirming the unreliability of prior evaluation standards. We also address noisy training data through an automated re-annotation pipeline, yielding TimeLens-100K, a large-scale, high-quality training dataset. Building on our data foundation, we conduct in-depth explorations of algorithmic design principles, yielding a series of meaningful insights and effective yet efficient practices. These include interleaved textual encoding for time representation, a thinking-free reinforcement learning with verifiable rewards (RLVR) approach as the training paradigm, and carefully designed recipes for RLVR training. These efforts culminate in TimeLens models, a family of MLLMs with state-of-the-art VTG performance among open-source models and even surpass proprietary models such as GPT-5 and Gemini-2.5-Flash. All codes, data, and models will be released to facilitate future research.

Yuying Ge, Annabella Macaluso, Li Erran Li et al.

Recent progress on vision-language foundation models have brought significant advancement to building general-purpose robots. By using the pre-trained models to encode the scene and instructions as inputs for decision making, the instruction-conditioned policy can generalize across different objects and tasks. While this is encouraging, the policy still fails in most cases given an unseen task or environment. In this work, we propose Policy Adaptation from Foundation model Feedback (PAFF). When deploying the trained policy to a new task or a new environment, we first let the policy play with randomly generated instructions to record the demonstrations. While the execution could be wrong, we can use the pre-trained foundation models to provide feedback to relabel the demonstrations. This automatically provides new pairs of demonstration-instruction data for policy fine-tuning. We evaluate our method on a broad range of experiments with the focus on generalization on unseen objects, unseen tasks, unseen environments, and sim-to-real transfer. We show PAFF improves baselines by a large margin in all cases. Our project page is available at https://geyuying.github.io/PAFF/

Yue Yang, Kaipeng Zhang, Yuying Ge et al.

Text-guided image generation has witnessed unprecedented progress due to the development of diffusion models. Beyond text and image, sound is a vital element within the sphere of human perception, offering vivid representations and naturally coinciding with corresponding scenes. Taking advantage of sound therefore presents a promising avenue for exploration within image generation research. However, the relationship between audio and image supervision remains significantly underdeveloped, and the scarcity of related, high-quality datasets brings further obstacles. In this paper, we propose a unified framework 'Align, Adapt, and Inject' (AAI) for sound-guided image generation, editing, and stylization. In particular, our method adapts input sound into a sound token, like an ordinary word, which can plug and play with existing powerful diffusion-based Text-to-Image (T2I) models. Specifically, we first train a multi-modal encoder to align audio representation with the pre-trained textual manifold and visual manifold, respectively. Then, we propose the audio adapter to adapt audio representation into an audio token enriched with specific semantics, which can be injected into a frozen T2I model flexibly. In this way, we are able to extract the dynamic information of varied sounds, while utilizing the formidable capability of existing T2I models to facilitate sound-guided image generation, editing, and stylization in a convenient and cost-effective manner. The experiment results confirm that our proposed AAI outperforms other text and sound-guided state-of-the-art methods. And our aligned multi-modal encoder is also competitive with other approaches in the audio-visual retrieval and audio-text retrieval tasks.

Yuying Ge, Sijie Zhao, Jinguo Zhu et al. · tencent-ai

The rapid evolution of multimodal foundation model has demonstrated significant progresses in vision-language understanding and generation, e.g., our previous work SEED-LLaMA. However, there remains a gap between its capability and the real-world applicability, primarily due to the model's limited capacity to effectively respond to various user instructions and interact with diverse visual data. In this work, we focus on bridging this gap through integrating two enhanced features: (1) comprehending images of arbitrary sizes and ratios, and (2) enabling multi-granularity image generation. We present a unified and versatile foundation model, namely, SEED-X, which is able to model multi-granularity visual semantics for comprehension and generation tasks. Besides the competitive results on public benchmarks, SEED-X demonstrates its effectiveness in handling real-world applications across various domains after instruction tuning. We hope that our work will inspire future research into what can be achieved by versatile multimodal foundation models in real-world applications. The models, codes, and datasets are released in https://github.com/AILab-CVC/SEED-X.

Bohao Li, Yuying Ge, Yi Chen et al. · tencent-ai

Comprehending text-rich visual content is paramount for the practical application of Multimodal Large Language Models (MLLMs), since text-rich scenarios are ubiquitous in the real world, which are characterized by the presence of extensive texts embedded within images. Recently, the advent of MLLMs with impressive versatility has raised the bar for what we can expect from MLLMs. However, their proficiency in text-rich scenarios has yet to be comprehensively and objectively assessed, since current MLLM benchmarks primarily focus on evaluating general visual comprehension. In this work, we introduce SEED-Bench-2-Plus, a benchmark specifically designed for evaluating \textbf{text-rich visual comprehension} of MLLMs. Our benchmark comprises 2.3K multiple-choice questions with precise human annotations, spanning three broad categories: Charts, Maps, and Webs, each of which covers a wide spectrum of text-rich scenarios in the real world. These categories, due to their inherent complexity and diversity, effectively simulate real-world text-rich environments. We further conduct a thorough evaluation involving 34 prominent MLLMs (including GPT-4V, Gemini-Pro-Vision and Claude-3-Opus) and emphasize the current limitations of MLLMs in text-rich visual comprehension. We hope that our work can serve as a valuable addition to existing MLLM benchmarks, providing insightful observations and inspiring further research in the area of text-rich visual comprehension with MLLMs. The dataset and evaluation code can be accessed at https://github.com/AILab-CVC/SEED-Bench.

Yuying Ge, Sijie Zhao, Chen Li et al. · tencent-ai

In this technical report, we introduce SEED-Data-Edit: a unique hybrid dataset for instruction-guided image editing, which aims to facilitate image manipulation using open-form language. SEED-Data-Edit is composed of three distinct types of data: (1) High-quality editing data produced by an automated pipeline, ensuring a substantial volume of diverse image editing pairs. (2) Real-world scenario data collected from the internet, which captures the intricacies of user intentions for promoting the practical application of image editing in the real world. (3) High-precision multi-turn editing data annotated by humans, which involves multiple rounds of edits for simulating iterative editing processes. The combination of these diverse data sources makes SEED-Data-Edit a comprehensive and versatile dataset for training language-guided image editing model. We fine-tune a pretrained Multimodal Large Language Model (MLLM) that unifies comprehension and generation with SEED-Data-Edit. The instruction tuned model demonstrates promising results, indicating the potential and effectiveness of SEED-Data-Edit in advancing the field of instructional image editing. The datasets are released in https://huggingface.co/datasets/AILab-CVC/SEED-Data-Edit.

Shengchao Zhou, Yuxin Chen, Yuying Ge et al.

Vision-language models (VLM) excel at general understanding yet remain weak at dynamic spatial reasoning (DSR), i.e., reasoning about the evolvement of object geometry and relationship in 3D space over time, largely due to the scarcity of scalable 4D-aware training resources. To bridge this gap across aspects of dataset, benchmark and model, we introduce DSR Suite. First, we propose an automated pipeline that generates multiple-choice question-answer pairs from in-the-wild videos for DSR. By leveraging modern vision foundation models, the pipeline extracts rich geometric and motion information, including camera poses, local point clouds, object masks, orientations, and 3D trajectories. These geometric cues enable the construction of DSR-Train for learning and further human-refined DSR-Bench for evaluation. Compared with previous works, our data emphasize (i) in-the-wild video sources, (ii) object- and scene-level 3D requirements, (iii) viewpoint transformations, (iv) multi-object interactions, and (v) fine-grained, procedural answers. Beyond data, we propose a lightweight Geometry Selection Module (GSM) to seamlessly integrate geometric priors into VLMs, which condenses question semantics and extracts question-relevant knowledge from pretrained 4D reconstruction priors into a compact set of geometry tokens. This targeted extraction avoids overwhelming the model with irrelevant knowledge. Experiments show that integrating DSR-Train and GSM into Qwen2.5-VL-7B significantly enhances its dynamic spatial reasoning capability, while maintaining accuracy on general video understanding benchmarks.

Haokun Lin, Teng Wang, Yixiao Ge et al.

Pioneering token-based works such as Chameleon and Emu3 have established a foundation for multimodal unification but face challenges of high training computational overhead and limited comprehension performance due to a lack of high-level semantics. In this paper, we introduce TokLIP, a visual tokenizer that enhances comprehension by semanticizing vector-quantized (VQ) tokens and incorporating CLIP-level semantics while enabling end-to-end multimodal autoregressive training with standard VQ tokens. TokLIP integrates a low-level discrete VQ tokenizer with a ViT-based token encoder to capture high-level continuous semantics. Unlike previous approaches (e.g., VILA-U) that discretize high-level features, TokLIP disentangles training objectives for comprehension and generation, allowing the direct application of advanced VQ tokenizers without the need for tailored quantization operations. Our empirical results demonstrate that TokLIP achieves exceptional data efficiency, empowering visual tokens with high-level semantic understanding while enhancing low-level generative capacity, making it well-suited for autoregressive Transformers in both comprehension and generation tasks. The code and models are available at https://github.com/TencentARC/TokLIP.

Junhao Cheng, Yuying Ge, Teng Wang et al.

Recent advances in CoT reasoning and RL post-training have been reported to enhance video reasoning capabilities of MLLMs. This progress naturally raises a question: can these models perform complex video reasoning in a manner comparable to human experts? However, existing video benchmarks primarily evaluate visual perception and grounding abilities, with questions that can be answered based on explicit prompts or isolated visual cues. Such benchmarks do not fully capture the intricacies of real-world reasoning, where humans must actively search for, integrate, and analyze multiple clues before reaching a conclusion. To address this issue, we present Video-Holmes, a benchmark inspired by the reasoning process of Sherlock Holmes, designed to evaluate the complex video reasoning capabilities of MLLMs. Video-Holmes consists of 1,837 questions derived from 270 manually annotated suspense short films, which spans seven carefully designed tasks. Each task is constructed by first identifying key events and causal relationships within films, and then designing questions that require models to actively locate and connect multiple relevant visual clues scattered across different video segments. Our comprehensive evaluation of state-of-the-art MLLMs reveals that, while these models generally excel at visual perception, they encounter substantial difficulties with integrating information and often miss critical clues. For example, the best-performing model, Gemini-2.5-Pro, achieves an accuracy of only 45%, with most models scoring below 40%. We aim that Video-Holmes can serve as a "Holmes-test" for multimodal reasoning, motivating models to reason more like humans and emphasizing the ongoing challenges in this field. The benchmark is released in https://github.com/TencentARC/Video-Holmes.

Lu Qiu, Yi Chen, Yuying Ge et al. · tencent-ai

The advent of Multimodal Large Language Models, leveraging the power of Large Language Models, has recently demonstrated superior multimodal understanding and reasoning abilities, heralding a new era for artificial general intelligence. However, achieving AGI necessitates more than just comprehension and reasoning. A crucial capability required is effective planning in diverse scenarios, which involves making reasonable decisions based on complex environments to solve real-world problems. Despite its importance, the planning abilities of current MLLMs in varied scenarios remain underexplored. In this paper, we introduce EgoPlan-Bench2, a rigorous and comprehensive benchmark designed to assess the planning capabilities of MLLMs across a wide range of real-world scenarios. EgoPlan-Bench2 encompasses everyday tasks spanning 4 major domains and 24 detailed scenarios, closely aligned with human daily life. EgoPlan-Bench2 is constructed through a semi-automatic process utilizing egocentric videos, complemented by manual verification. Grounded in a first-person perspective, it mirrors the way humans approach problem-solving in everyday life. We evaluate 21 competitive MLLMs and provide an in-depth analysis of their limitations, revealing that they face significant challenges in real-world planning. To further improve the planning proficiency of current MLLMs, we propose a training-free approach using multimodal Chain-of-Thought (CoT) prompting through investigating the effectiveness of various multimodal prompts in complex planning. Our approach enhances the performance of GPT-4V by 10.24 on EgoPlan-Bench2 without additional training. Our work not only sheds light on the current limitations of MLLMs in planning, but also provides insights for future enhancements in this critical area. We have made data and code available at https://qiulu66.github.io/egoplanbench2/.

Junhao Cheng, Yuying Ge, Yixiao Ge et al.

Recent advancements in image and video synthesis have opened up new promise in generative games. One particularly intriguing application is transforming characters from anime films into interactive, playable entities. This allows players to immerse themselves in the dynamic anime world as their favorite characters for life simulation through language instructions. Such games are defined as infinite game since they eliminate predetermined boundaries and fixed gameplay rules, where players can interact with the game world through open-ended language and experience ever-evolving storylines and environments. Recently, a pioneering approach for infinite anime life simulation employs large language models (LLMs) to translate multi-turn text dialogues into language instructions for image generation. However, it neglects historical visual context, leading to inconsistent gameplay. Furthermore, it only generates static images, failing to incorporate the dynamics necessary for an engaging gaming experience. In this work, we propose AnimeGamer, which is built upon Multimodal Large Language Models (MLLMs) to generate each game state, including dynamic animation shots that depict character movements and updates to character states, as illustrated in Figure 1. We introduce novel action-aware multimodal representations to represent animation shots, which can be decoded into high-quality video clips using a video diffusion model. By taking historical animation shot representations as context and predicting subsequent representations, AnimeGamer can generate games with contextual consistency and satisfactory dynamics. Extensive evaluations using both automated metrics and human evaluations demonstrate that AnimeGamer outperforms existing methods in various aspects of the gaming experience. Codes and checkpoints are available at https://github.com/TencentARC/AnimeGamer.

Yuxin Guo, Teng Wang, Yuying Ge et al.

Recent advances in text-to-audio (TTA) generation excel at synthesizing short audio clips but struggle with long-form narrative audio, which requires temporal coherence and compositional reasoning. To address this gap, we propose AudioStory, a unified framework that integrates large language models (LLMs) with TTA systems to generate structured, long-form audio narratives. AudioStory possesses strong instruction-following reasoning generation capabilities. It employs LLMs to decompose complex narrative queries into temporally ordered sub-tasks with contextual cues, enabling coherent scene transitions and emotional tone consistency. AudioStory has two appealing features: (1) Decoupled bridging mechanism: AudioStory disentangles LLM-diffuser collaboration into two specialized components, i.e., a bridging query for intra-event semantic alignment and a residual query for cross-event coherence preservation. (2) End-to-end training: By unifying instruction comprehension and audio generation within a single end-to-end framework, AudioStory eliminates the need for modular training pipelines while enhancing synergy between components. Furthermore, we establish a benchmark AudioStory-10K, encompassing diverse domains such as animated soundscapes and natural sound narratives. Extensive experiments show the superiority of AudioStory on both single-audio generation and narrative audio generation, surpassing prior TTA baselines in both instruction-following ability and audio fidelity. Our code is available at https://github.com/TencentARC/AudioStory

Yatai Ji, Teng Wang, Yuying Ge et al.

Discrete diffusion models have emerged as a promising direction for vision-language tasks, offering bidirectional context modeling and theoretical parallelization. However, their practical application is severely hindered by a train-inference discrepancy, which leads to catastrophic error cascades: initial token errors during parallel decoding pollute the generation context, triggering a chain reaction of compounding errors and leading to syntactic errors and semantic hallucinations. To address this fundamental challenge, we reframe the generation process from passive denoising to active refining. We introduce ReDiff, a refining-enhanced diffusion framework that teaches the model to identify and correct its own errors. Our approach features a two-stage training process: first, we instill a foundational revision capability by training the model to revise synthetic errors; second, we implement a novel online self-correction loop where the model is explicitly trained to revise its own flawed drafts by learning from an expert's corrections. This mistake-driven learning endows the model with the crucial ability to revisit and refine its already generated output, effectively breaking the error cascade. Extensive experiments demonstrate that ReDiff significantly improves the coherence and factual accuracy of generated content, enabling stable and efficient parallel generation far superior to traditional denoising methods. Our codes and models are available at https://rediff-hku.github.io/.

Jinguo Zhu, Xiaohan Ding, Yixiao Ge et al. · tencent-ai

In this work, we introduce Vision-Language Generative Pre-trained Transformer (VL-GPT), a transformer model proficient at concurrently perceiving and generating visual and linguistic data. VL-GPT achieves a unified pre-training approach for both image and text modalities by employing a straightforward auto-regressive objective, thereby enabling the model to process image and text as seamlessly as a language model processes text. To accomplish this, we initially propose a novel image tokenizer-detokenizer framework for visual data, specifically designed to transform raw images into a sequence of continuous embeddings and reconstruct them accordingly. In combination with the existing text tokenizer and detokenizer, this framework allows for the encoding of interleaved image-text data into a multimodal sequence, which can subsequently be fed into the transformer model. Consequently, VL-GPT can perform large-scale pre-training on multimodal corpora utilizing a unified auto-regressive objective (i.e., next-token prediction). Upon completion of pre-training, VL-GPT exhibits remarkable zero-shot and few-shot performance across a diverse range of vision and language understanding and generation tasks, including image captioning, visual question answering, text-to-image generation, and more. Additionally, the pre-trained model retrains in-context learning capabilities when provided with multimodal prompts. We further conduct instruction tuning on our VL-GPT, highlighting its exceptional potential for multimodal assistance. The source code and model weights shall be released.

Yi Chen, Yuying Ge, Yixiao Ge et al. · tencent-ai

The pursuit of artificial general intelligence (AGI) has been accelerated by Multimodal Large Language Models (MLLMs), which exhibit superior reasoning, generalization capabilities, and proficiency in processing multimodal inputs. A crucial milestone in the evolution of AGI is the attainment of human-level planning, a fundamental ability for making informed decisions in complex environments, and solving a wide range of real-world problems. Despite the impressive advancements in MLLMs, a question remains: How far are current MLLMs from achieving human-level planning? To shed light on this question, we introduce EgoPlan-Bench, a comprehensive benchmark to evaluate the planning abilities of MLLMs in real-world scenarios from an egocentric perspective, mirroring human perception. EgoPlan-Bench emphasizes the evaluation of planning capabilities of MLLMs, featuring realistic tasks, diverse action plans, and intricate visual observations. Our rigorous evaluation of a wide range of MLLMs reveals that EgoPlan-Bench poses significant challenges, highlighting a substantial scope for improvement in MLLMs to achieve human-level task planning. To facilitate this advancement, we further present EgoPlan-IT, a specialized instruction-tuning dataset that effectively enhances model performance on EgoPlan-Bench. We have made all codes, data, and a maintained benchmark leaderboard available to advance future research.

Yi Chen, Yuying Ge, Rui Wang et al.

Recent reinforcement learning approaches, such as outcome-supervised GRPO, have advanced Chain-of-Thought reasoning in large language models (LLMs), yet their adaptation to multimodal LLMs (MLLMs) is unexplored. To address the lack of rigorous evaluation for MLLM post-training methods, we introduce SEED-Bench-R1, a benchmark with complex real-world videos requiring balanced perception and reasoning. It offers a large training set and evaluates generalization across three escalating challenges: in-distribution, cross-environment, and cross-environment-task scenarios. Using SEED-Bench-R1, we find that standard GRPO, while improving answer accuracy, often reduces logical coherence between reasoning steps and answers, with only a 57.9% consistency rate. This stems from reward signals focusing solely on final answers, encouraging shortcuts, and strict KL penalties limiting exploration.To address this, we propose GRPO-CARE, a consistency-aware RL framework optimizing both answer correctness and reasoning coherence without explicit supervision. GRPO-CARE introduces a two-tiered reward: (1) a base reward for answer correctness, and (2) an adaptive consistency bonus, computed by comparing the model's reasoning-to-answer likelihood (via a slowly-evolving reference model) against group peers.This dual mechanism amplifies rewards for reasoning paths that are both correct and logically consistent. Replacing KL penalties with this adaptive bonus, GRPO-CARE outperforms standard GRPO on SEED-Bench-R1, achieving a 6.7% performance gain on the hardest evaluation level and a 24.5% improvement in consistency. It also shows strong transferability, improving model performance across diverse video understanding benchmarks. Our work contributes a systematically designed benchmark and a generalizable post-training framework, advancing the development of more interpretable and robust MLLMs.

Yi Chen, Yuying Ge, Rui Wang et al.

Recent advancements in Chain of Thought (COT) generation have significantly improved the reasoning capabilities of Large Language Models (LLMs), with reinforcement learning (RL) emerging as an effective post-training approach. Multimodal Large Language Models (MLLMs) inherit this reasoning potential but remain underexplored in tasks requiring both perception and logical reasoning. To address this, we introduce SEED-Bench-R1, a benchmark designed to systematically evaluate post-training methods for MLLMs in video understanding. It includes intricate real-world videos and complex everyday planning tasks in the format of multiple-choice questions, requiring sophisticated perception and reasoning. SEED-Bench-R1 assesses generalization through a three-level hierarchy: in-distribution, cross-environment, and cross-environment-task scenarios, equipped with a large-scale training dataset with easily verifiable ground-truth answers. Using Qwen2-VL-Instruct-7B as a base model, we compare RL with supervised fine-tuning (SFT), demonstrating RL's data efficiency and superior performance on both in-distribution and out-of-distribution tasks, even outperforming SFT on general video understanding benchmarks like LongVideoBench. Our detailed analysis reveals that RL enhances visual perception but often produces less logically coherent reasoning chains. We identify key limitations such as inconsistent reasoning and overlooked visual cues, and suggest future improvements in base model reasoning, reward modeling, and RL robustness against noisy signals.

Boyu Chen, Yi Chen, Lu Qiu et al.

Scaling humanoid foundation models is bottlenecked by the scarcity of robotic data. While massive egocentric human data offers a scalable alternative, bridging the cross-embodiment chasm remains a fundamental challenge due to kinematic mismatches. We introduce UniT (Unified Latent Action Tokenizer via Visual Anchoring), a framework that establishes a unified physical language for human-to-humanoid transfer. Grounded in the philosophy that heterogeneous kinematics share universal visual consequences, UniT employs a tri-branch cross-reconstruction mechanism: actions predict vision to anchor kinematics to physical outcomes, while vision reconstructs actions to filter out irrelevant visual confounders. Concurrently, a fusion branch synergies these purified modalities into a shared discrete latent space of embodiment-agnostic physical intents. We validate UniT across two paradigms: 1) Policy Learning (VLA-UniT): By predicting these unified tokens, it effectively leverages diverse human data to achieve state-of-the-art data efficiency and robust out-of-distribution (OOD) generalization on both humanoid simulation benchmark and real-world deployments, notably demonstrating zero-shot task transfer. 2) World Modeling (WM-UniT): By aligning cross-embodiment dynamics via unified tokens as conditions, it realizes direct human-to-humanoid action transfer. This alignment ensures that human data seamlessly translates into enhanced action controllability for humanoid video generation. Ultimately, by inducing a highly aligned cross-embodiment representation (empirically verified by t-SNE visualizations revealing the convergence of human and humanoid features into a shared manifold), UniT offers a scalable path to distill vast human knowledge into general-purpose humanoid capabilities.

Yi Chen, Yuying Ge, Weiliang Tang et al.

Recent developments in Large Language Models pre-trained on extensive corpora have shown significant success in various natural language processing tasks with minimal fine-tuning. This success offers new promise for robotics, which has long been constrained by the high cost of action-labeled data. We ask: given the abundant video data containing interaction-related knowledge available as a rich "corpus", can a similar generative pre-training approach be effectively applied to enhance robot learning? The key challenge is to identify an effective representation for autoregressive pre-training that benefits robot manipulation tasks. Inspired by the way humans learn new skills through observing dynamic environments, we propose that effective robotic learning should emphasize motion-related knowledge, which is closely tied to low-level actions and is hardware-agnostic, facilitating the transfer of learned motions to actual robot actions. To this end, we introduce Moto, which converts video content into latent Motion Token sequences by a Latent Motion Tokenizer, learning a bridging "language" of motion from videos in an unsupervised manner. We pre-train Moto-GPT through motion token autoregression, enabling it to capture diverse visual motion knowledge. After pre-training, Moto-GPT demonstrates the promising ability to produce semantically interpretable motion tokens, predict plausible motion trajectories, and assess trajectory rationality through output likelihood. To transfer learned motion priors to real robot actions, we implement a co-fine-tuning strategy that seamlessly bridges latent motion token prediction and real robot control. Extensive experiments show that the fine-tuned Moto-GPT exhibits superior robustness and efficiency on robot manipulation benchmarks, underscoring its effectiveness in transferring knowledge from video data to downstream visual manipulation tasks.

Yuying Ge, Yixiao Ge, Chen Li et al.

Real-world user-generated short videos, especially those distributed on platforms such as WeChat Channel and TikTok, dominate the mobile internet. However, current large multimodal models lack essential temporally-structured, detailed, and in-depth video comprehension capabilities, which are the cornerstone of effective video search and recommendation, as well as emerging video applications. Understanding real-world shorts is actually challenging due to their complex visual elements, high information density in both visuals and audio, and fast pacing that focuses on emotional expression and viewpoint delivery. This requires advanced reasoning to effectively integrate multimodal information, including visual, audio, and text. In this work, we introduce ARC-Hunyuan-Video, a multimodal model that processes visual, audio, and textual signals from raw video inputs end-to-end for structured comprehension. The model is capable of multi-granularity timestamped video captioning and summarization, open-ended video question answering, temporal video grounding, and video reasoning. Leveraging high-quality data from an automated annotation pipeline, our compact 7B-parameter model is trained through a comprehensive regimen: pre-training, instruction fine-tuning, cold start, reinforcement learning (RL) post-training, and final instruction fine-tuning. Quantitative evaluations on our introduced benchmark ShortVid-Bench and qualitative comparisons demonstrate its strong performance in real-world video comprehension, and it supports zero-shot or fine-tuning with a few samples for diverse downstream applications. The real-world production deployment of our model has yielded tangible and measurable improvements in user engagement and satisfaction, a success supported by its remarkable efficiency, with stress tests indicating an inference time of just 10 seconds for a one-minute video on H20 GPU.

Shijie Ma, Yuying Ge, Teng Wang et al.

The synergy between generative and discriminative models receives growing attention. While discriminative Contrastive Language-Image Pre-Training (CLIP) excels in high-level semantics, it struggles with perceiving fine-grained visual details. Generally, to enhance representations, generative models take CLIP's visual features as conditions for reconstruction. However, the underlying principle remains underexplored. In this work, we empirically found that visually perfect generations are not always optimal for representation enhancement. The essence lies in effectively extracting fine-grained knowledge from generative models while mitigating irrelevant information. To explore critical factors, we delve into three aspects: (1) Conditioning mechanisms: We found that even a small number of local tokens can drastically reduce the difficulty of reconstruction, leading to collapsed training. We thus conclude that utilizing only global visual tokens as conditions is the most effective strategy. (2) Denoising configurations: We observed that end-to-end training introduces extraneous information. To address this, we propose a two-stage training strategy to prioritize learning useful visual knowledge. Additionally, we demonstrate that lightweight denoisers can yield remarkable improvements. (3) Generation paradigms: We explore both continuous and discrete denoisers with desirable outcomes, validating the versatility of our method. Through our in-depth explorations, we have finally arrived at an effective method, namely GenHancer, which consistently outperforms prior arts on the MMVP-VLM benchmark, e.g., 6.0% on OpenAICLIP. The enhanced CLIP can be further plugged into multimodal large language models for better vision-centric performance. All the models and codes are made publicly available.

Yuying Ge, Yizhuo Li, Yixiao Ge et al. · tencent-ai

In recent years, there has been a significant surge of interest in unifying image comprehension and generation within Large Language Models (LLMs). This growing interest has prompted us to explore extending this unification to videos. The core challenge lies in developing a versatile video tokenizer that captures both the spatial characteristics and temporal dynamics of videos to obtain representations for LLMs, and the representations can be further decoded into realistic video clips to enable video generation. In this work, we introduce Divot, a Diffusion-Powered Video Tokenizer, which leverages the diffusion process for self-supervised video representation learning. We posit that if a video diffusion model can effectively de-noise video clips by taking the features of a video tokenizer as the condition, then the tokenizer has successfully captured robust spatial and temporal information. Additionally, the video diffusion model inherently functions as a de-tokenizer, decoding videos from their representations. Building upon the Divot tokenizer, we present Divot-Vicuna through video-to-text autoregression and text-to-video generation by modeling the distributions of continuous-valued Divot features with a Gaussian Mixture Model. Experimental results demonstrate that our diffusion-based video tokenizer, when integrated with a pre-trained LLM, achieves competitive performance across various video comprehension and generation benchmarks. The instruction tuned Divot-Vicuna also excels in video storytelling, generating interleaved narratives and corresponding videos.

Yi Chen, Yuying Ge, Hui Zhou et al.

The development of Vision-Language-Action (VLA) models has been significantly accelerated by pre-trained Vision-Language Models (VLMs). However, most existing end-to-end VLAs treat the VLM primarily as a multimodal encoder, directly mapping vision-language features to low-level actions. This paradigm underutilizes the VLM's potential in high-level decision making and introduces training instability, frequently degrading its rich semantic representations. To address these limitations, we introduce DIAL, a framework bridging high-level decision making and low-level motor execution through a differentiable latent intent bottleneck. Specifically, a VLM-based System-2 performs latent world modeling by synthesizing latent visual foresight within the VLM's native feature space; this foresight explicitly encodes intent and serves as the structural bottleneck. A lightweight System-1 policy then decodes this predicted intent together with the current observation into precise robot actions via latent inverse dynamics. To ensure optimization stability, we employ a two-stage training paradigm: a decoupled warmup phase where System-2 learns to predict latent futures while System-1 learns motor control under ground-truth future guidance within a unified feature space, followed by seamless end-to-end joint optimization. This enables action-aware gradients to refine the VLM backbone in a controlled manner, preserving pre-trained knowledge. Extensive experiments on the RoboCasa GR1 Tabletop benchmark show that DIAL establishes a new state-of-the-art, achieving superior performance with 10x fewer demonstrations than prior methods. Furthermore, by leveraging heterogeneous human demonstrations, DIAL learns physically grounded manipulation priors and exhibits robust zero-shot generalization to unseen objects and novel configurations during real-world deployment on a humanoid robot.

Yizhuo Li, Yuying Ge, Yixiao Ge et al.

This paper presents a novel framework for aligning learnable latent spaces to arbitrary target distributions by leveraging flow-based generative models as priors. Our method first pretrains a flow model on the target features to capture the underlying distribution. This fixed flow model subsequently regularizes the latent space via an alignment loss, which reformulates the flow matching objective to treat the latents as optimization targets. We formally prove that minimizing this alignment loss establishes a computationally tractable surrogate objective for maximizing a variational lower bound on the log-likelihood of latents under the target distribution. Notably, the proposed method eliminates computationally expensive likelihood evaluations and avoids ODE solving during optimization. As a proof of concept, we demonstrate in a controlled setting that the alignment loss landscape closely approximates the negative log-likelihood of the target distribution. We further validate the effectiveness of our approach through large-scale image generation experiments on ImageNet with diverse target distributions, accompanied by detailed discussions and ablation studies. With both theoretical and empirical validation, our framework paves a new way for latent space alignment.

Yizhuo Li, Yuying Ge, Yixiao Ge et al. · tencent-ai

Videos are inherently temporal sequences by their very nature. In this work, we explore the potential of modeling videos in a chronological and scalable manner with autoregressive (AR) language models, inspired by their success in natural language processing. We introduce DiCoDe, a novel approach that leverages Diffusion-Compressed Deep Tokens to generate videos with a language model in an autoregressive manner. Unlike existing methods that employ low-level representations with limited compression rates, DiCoDe utilizes deep tokens with a considerable compression rate (a 1000x reduction in token count). This significant compression is made possible by a tokenizer trained through leveraging the prior knowledge of video diffusion models. Deep tokens enable DiCoDe to employ vanilla AR language models for video generation, akin to translating one visual "language" into another. By treating videos as temporal sequences, DiCoDe fully harnesses the capabilities of language models for autoregressive generation. DiCoDe is scalable using readily available AR architectures, and is capable of generating videos ranging from a few seconds to one minute using only 4 A100 GPUs for training. We evaluate DiCoDe both quantitatively and qualitatively, demonstrating that it performs comparably to existing methods in terms of quality while ensuring efficient training. To showcase its scalability, we release a series of DiCoDe configurations with varying parameter sizes and observe a consistent improvement in performance as the model size increases from 100M to 3B. We believe that DiCoDe's exploration in academia represents a promising initial step toward scalable video modeling with AR language models, paving the way for the development of larger and more powerful video generation models.

Junfu Pu, Teng Wang, Yixiao Ge et al.

The proliferation of hour-long videos (e.g., lectures, podcasts, documentaries) has intensified demand for efficient content structuring. However, existing approaches are constrained by small-scale training with annotations that are typical short and coarse, restricting generalization to nuanced transitions in long videos. We introduce ARC-Chapter, the first large-scale video chaptering model trained on over million-level long video chapters, featuring bilingual, temporally grounded, and hierarchical chapter annotations. To achieve this goal, we curated a bilingual English-Chinese chapter dataset via a structured pipeline that unifies ASR transcripts, scene texts, visual captions into multi-level annotations, from short title to long summaries. We demonstrate clear performance improvements with data scaling, both in data volume and label intensity. Moreover, we design a new evaluation metric termed GRACE, which incorporates many-to-one segment overlaps and semantic similarity, better reflecting real-world chaptering flexibility. Extensive experiments demonstrate that ARC-Chapter establishes a new state-of-the-art by a significant margin, outperforming the previous best by 14.0% in F1 score and 11.3% in SODA score. Moreover, ARC-Chapter shows excellent transferability, improving the state-of-the-art on downstream tasks like dense video captioning on YouCook2.

Lu Qiu, Yizhuo Li, Yuying Ge et al.

Recent advances in AI-generated content (AIGC) have significantly accelerated animation production. To produce engaging animations, it is essential to generate coherent multi-shot video clips with narrative scripts and character references. However, existing public datasets primarily focus on real-world scenarios with global descriptions, and lack reference images for consistent character guidance. To bridge this gap, we present AnimeShooter, a reference-guided multi-shot animation dataset. AnimeShooter features comprehensive hierarchical annotations and strong visual consistency across shots through an automated pipeline. Story-level annotations provide an overview of the narrative, including the storyline, key scenes, and main character profiles with reference images, while shot-level annotations decompose the story into consecutive shots, each annotated with scene, characters, and both narrative and descriptive visual captions. Additionally, a dedicated subset, AnimeShooter-audio, offers synchronized audio tracks for each shot, along with audio descriptions and sound sources. To demonstrate the effectiveness of AnimeShooter and establish a baseline for the reference-guided multi-shot video generation task, we introduce AnimeShooterGen, which leverages Multimodal Large Language Models (MLLMs) and video diffusion models. The reference image and previously generated shots are first processed by MLLM to produce representations aware of both reference and context, which are then used as the condition for the diffusion model to decode the subsequent shot. Experimental results show that the model trained on AnimeShooter achieves superior cross-shot visual consistency and adherence to reference visual guidance, which highlight the value of our dataset for coherent animated video generation.

Xiaohu Jiang, Yixiao Ge, Yuying Ge et al.

Image-text training like CLIP has dominated the pretraining of vision foundation models in recent years. Subsequent efforts have been made to introduce region-level visual learning into CLIP's pretraining but face scalability challenges due to the lack of large-scale region-level datasets. Drawing inspiration from supervised fine-tuning (SFT) in natural language processing such as instruction tuning, we explore the potential of fine-grained SFT in enhancing the generation of vision foundation models after their pretraining. Thus a two-stage method ViSFT (Vision SFT) is proposed to unleash the fine-grained knowledge of vision foundation models. In ViSFT, the vision foundation model is enhanced by performing visual joint learning on some in-domain tasks and then tested on out-of-domain benchmarks. With updating using ViSFT on 8 V100 GPUs in less than 2 days, a vision transformer with over 4.4B parameters shows improvements across various out-of-domain benchmarks including vision and vision-linguistic scenarios.

Yuying Ge, Yibing Song, Ruimao Zhang et al.

Dancing video retargeting aims to synthesize a video that transfers the dance movements from a source video to a target person. Previous work need collect a several-minute-long video of a target person with thousands of frames to train a personalized model. However, the trained model can only generate videos of the same person. To address the limitations, recent work tackled few-shot dancing video retargeting, which learns to synthesize videos of unseen persons by leveraging a few frames of them. In practice, given a few frames of a person, these work simply regarded them as a batch of individual images without temporal correlations, thus generating temporally incoherent dancing videos of low visual quality. In this work, we model a few frames of a person as a series of dancing moves, where each move contains two consecutive frames, to extract the appearance patterns and the temporal dynamics of this person. We propose MetaDance, which utilizes temporal-aware meta-learning to optimize the initialization of a model through the synthesis of dancing moves, such that the meta-trained model can be efficiently tuned towards enhanced visual quality and strengthened temporal stability for unseen persons with a few frames. Extensive evaluations show large superiority of our method.

Yuying Ge, Yixiao Ge, Xihui Liu et al.

Pre-training a model to learn transferable video-text representation for retrieval has attracted a lot of attention in recent years. Previous dominant works mainly adopt two separate encoders for efficient retrieval, but ignore local associations between videos and texts. Another line of research uses a joint encoder to interact video with texts, but results in low efficiency since each text-video pair needs to be fed into the model. In this work, we enable fine-grained video-text interactions while maintaining high efficiency for retrieval via a novel pretext task, dubbed as Multiple Choice Questions (MCQ), where a parametric module BridgeFormer is trained to answer the "questions" constructed by the text features via resorting to the video features. Specifically, we exploit the rich semantics of text (i.e., nouns and verbs) to build questions, with which the video encoder can be trained to capture more regional content and temporal dynamics. In the form of questions and answers, the semantic associations between local video-text features can be properly established. BridgeFormer is able to be removed for downstream retrieval, rendering an efficient and flexible model with only two encoders. Our method outperforms state-of-the-art methods on the popular text-to-video retrieval task in five datasets with different experimental setups (i.e., zero-shot and fine-tune), including HowTo100M (one million videos). We further conduct zero-shot action recognition, which can be cast as video-to-text retrieval, and our approach also significantly surpasses its counterparts. As an additional benefit, our method achieves competitive results with much shorter pre-training videos on single-modality downstream tasks, e.g., action recognition with linear evaluation.

Yuying Ge, Ruimao Zhang, Ping Luo

Recent advanced methods for fashion landmark detection are mainly driven by training convolutional neural networks on large-scale fashion datasets, which has a large number of annotated landmarks. However, such large-scale annotations are difficult and expensive to obtain in real-world applications, thus models that can generalize well from a small amount of labelled data are desired. We investigate this problem of few-shot fashion landmark detection, where only a few labelled samples are available for an unseen task. This work proposes a novel framework named MetaCloth via meta-learning, which is able to learn unseen tasks of dense fashion landmark detection with only a few annotated samples. Unlike previous meta-learning work that focus on solving "N-way K-shot" tasks, where each task predicts N number of classes by training with K annotated samples for each class (N is fixed for all seen and unseen tasks), a task in MetaCloth detects N different landmarks for different clothing categories using K samples, where N varies across tasks, because different clothing categories usually have various number of landmarks. Therefore, numbers of parameters are various for different seen and unseen tasks in MetaCloth. MetaCloth is carefully designed to dynamically generate different numbers of parameters for different tasks, and learn a generalizable feature extraction network from a few annotated samples with a set of good initialization parameters. Extensive experiments show that MetaCloth outperforms its counterparts by a large margin.

Chongjian Ge, Yibing Song, Yuying Ge et al.

Image virtual try-on replaces the clothes on a person image with a desired in-shop clothes image. It is challenging because the person and the in-shop clothes are unpaired. Existing methods formulate virtual try-on as either in-painting or cycle consistency. Both of these two formulations encourage the generation networks to reconstruct the input image in a self-supervised manner. However, existing methods do not differentiate clothing and non-clothing regions. A straight-forward generation impedes virtual try-on quality because of the heavily coupled image contents. In this paper, we propose a Disentangled Cycle-consistency Try-On Network (DCTON). The DCTON is able to produce highly-realistic try-on images by disentangling important components of virtual try-on including clothes warping, skin synthesis, and image composition. To this end, DCTON can be naturally trained in a self-supervised manner following cycle consistency learning. Extensive experiments on challenging benchmarks show that DCTON outperforms state-of-the-art approaches favorably.

Yuying Ge, Yibing Song, Ruimao Zhang et al.

Image virtual try-on aims to fit a garment image (target clothes) to a person image. Prior methods are heavily based on human parsing. However, slightly-wrong segmentation results would lead to unrealistic try-on images with large artifacts. Inaccurate parsing misleads parser-based methods to produce visually unrealistic results where artifacts usually occur. A recent pioneering work employed knowledge distillation to reduce the dependency of human parsing, where the try-on images produced by a parser-based method are used as supervisions to train a "student" network without relying on segmentation, making the student mimic the try-on ability of the parser-based model. However, the image quality of the student is bounded by the parser-based model. To address this problem, we propose a novel approach, "teacher-tutor-student" knowledge distillation, which is able to produce highly photo-realistic images without human parsing, possessing several appealing advantages compared to prior arts. (1) Unlike existing work, our approach treats the fake images produced by the parser-based method as "tutor knowledge", where the artifacts can be corrected by real "teacher knowledge", which is extracted from the real person images in a self-supervised way. (2) Other than using real images as supervisions, we formulate knowledge distillation in the try-on problem as distilling the appearance flows between the person image and the garment image, enabling us to find accurate dense correspondences between them to produce high-quality results. (3) Extensive evaluations show large superiority of our method (see Fig. 1).

Yuying Ge, Ruimao Zhang, Lingyun Wu et al.

Understanding fashion images has been advanced by benchmarks with rich annotations such as DeepFashion, whose labels include clothing categories, landmarks, and consumer-commercial image pairs. However, DeepFashion has nonnegligible issues such as single clothing-item per image, sparse landmarks (4~8 only), and no per-pixel masks, making it had significant gap from real-world scenarios. We fill in the gap by presenting DeepFashion2 to address these issues. It is a versatile benchmark of four tasks including clothes detection, pose estimation, segmentation, and retrieval. It has 801K clothing items where each item has rich annotations such as style, scale, viewpoint, occlusion, bounding box, dense landmarks and masks. There are also 873K Commercial-Consumer clothes pairs. A strong baseline is proposed, called Match R-CNN, which builds upon Mask R-CNN to solve the above four tasks in an end-to-end manner. Extensive evaluations are conducted with different criterions in DeepFashion2.

Ruimao Zhang, Hongbin Sun, Jingyu Li et al.

Video person re-identification attracts much attention in recent years. It aims to match image sequences of pedestrians from different camera views. Previous approaches usually improve this task from three aspects, including a) selecting more discriminative frames, b) generating more informative temporal representations, and c) developing more effective distance metrics. To address the above issues, we present a novel and practical deep architecture for video person re-identification termed Self-and-Collaborative Attention Network (SCAN). It has several appealing properties. First, SCAN adopts non-parametric attention mechanism to refine the intra-sequence and inter-sequence feature representation of videos, and outputs self-and-collaborative feature representation for each video, making the discriminative frames aligned between the probe and gallery sequences.Second, beyond existing models, a generalized pairwise similarity measurement is proposed to calculate the similarity feature representations of video pairs, enabling computing the matching scores by the binary classifier. Third, a dense clip segmentation strategy is also introduced to generate rich probe-gallery pairs to optimize the model. Extensive experiments demonstrate the effectiveness of SCAN, which outperforms the best-performing baselines on iLIDS-VID, PRID2011 and MARS dataset, respectively.