Lianghua Huang

Zhantao Yang, Ruili Feng, Han Zhang et al. · tsinghua

Diffusion models, which employ stochastic differential equations to sample images through integrals, have emerged as a dominant class of generative models. However, the rationality of the diffusion process itself receives limited attention, leaving the question of whether the problem is well-posed and well-conditioned. In this paper, we explore a perplexing tendency of diffusion models: they often display the infinite Lipschitz property of the network with respect to time variable near the zero point. We provide theoretical proofs to illustrate the presence of infinite Lipschitz constants and empirical results to confirm it. The Lipschitz singularities pose a threat to the stability and accuracy during both the training and inference processes of diffusion models. Therefore, the mitigation of Lipschitz singularities holds great potential for enhancing the performance of diffusion models. To address this challenge, we propose a novel approach, dubbed E-TSDM, which alleviates the Lipschitz singularities of the diffusion model near the zero point of timesteps. Remarkably, our technique yields a substantial improvement in performance. Moreover, as a byproduct of our method, we achieve a dramatic reduction in the Fréchet Inception Distance of acceleration methods relying on network Lipschitz, including DDIM and DPM-Solver, by over 33%. Extensive experiments on diverse datasets validate our theory and method. Our work may advance the understanding of the general diffusion process, and also provide insights for the design of diffusion models.

Lianghua Huang, Di Chen, Yu Liu et al.

Recent large-scale generative models learned on big data are capable of synthesizing incredible images yet suffer from limited controllability. This work offers a new generation paradigm that allows flexible control of the output image, such as spatial layout and palette, while maintaining the synthesis quality and model creativity. With compositionality as the core idea, we first decompose an image into representative factors, and then train a diffusion model with all these factors as the conditions to recompose the input. At the inference stage, the rich intermediate representations work as composable elements, leading to a huge design space (i.e., exponentially proportional to the number of decomposed factors) for customizable content creation. It is noteworthy that our approach, which we call Composer, supports various levels of conditions, such as text description as the global information, depth map and sketch as the local guidance, color histogram for low-level details, etc. Besides improving controllability, we confirm that Composer serves as a general framework and facilitates a wide range of classical generative tasks without retraining. Code and models will be made available.

Xi Chen, Lianghua Huang, Yu Liu et al.

This work presents AnyDoor, a diffusion-based image generator with the power to teleport target objects to new scenes at user-specified locations in a harmonious way. Instead of tuning parameters for each object, our model is trained only once and effortlessly generalizes to diverse object-scene combinations at the inference stage. Such a challenging zero-shot setting requires an adequate characterization of a certain object. To this end, we complement the commonly used identity feature with detail features, which are carefully designed to maintain texture details yet allow versatile local variations (e.g., lighting, orientation, posture, etc.), supporting the object in favorably blending with different surroundings. We further propose to borrow knowledge from video datasets, where we can observe various forms (i.e., along the time axis) of a single object, leading to stronger model generalizability and robustness. Extensive experiments demonstrate the superiority of our approach over existing alternatives as well as its great potential in real-world applications, such as virtual try-on and object moving. Project page is https://damo-vilab.github.io/AnyDoor-Page/.

Han Zhang, Ruili Feng, Zhantao Yang et al.

Diffusion models, which learn to reverse a signal destruction process to generate new data, typically require the signal at each step to have the same dimension. We argue that, considering the spatial redundancy in image signals, there is no need to maintain a high dimensionality in the evolution process, especially in the early generation phase. To this end, we make a theoretical generalization of the forward diffusion process via signal decomposition. Concretely, we manage to decompose an image into multiple orthogonal components and control the attenuation of each component when perturbing the image. That way, along with the noise strength increasing, we are able to diminish those inconsequential components and thus use a lower-dimensional signal to represent the source, barely losing information. Such a reformulation allows to vary dimensions in both training and inference of diffusion models. Extensive experiments on a range of datasets suggest that our approach substantially reduces the computational cost and achieves on-par or even better synthesis performance compared to baseline methods. We also show that our strategy facilitates high-resolution image synthesis and improves FID of diffusion model trained on FFHQ at $1024\times1024$ resolution from 52.40 to 10.46. Code and models will be made publicly available.

Shiyue Cao, Yueqin Yin, Lianghua Huang et al.

Vector-quantized image modeling has shown great potential in synthesizing high-quality images. However, generating high-resolution images remains a challenging task due to the quadratic computational overhead of the self-attention process. In this study, we seek to explore a more efficient two-stage framework for high-resolution image generation with improvements in the following three aspects. (1) Based on the observation that the first quantization stage has solid local property, we employ a local attention-based quantization model instead of the global attention mechanism used in previous methods, leading to better efficiency and reconstruction quality. (2) We emphasize the importance of multi-grained feature interaction during image generation and introduce an efficient attention mechanism that combines global attention (long-range semantic consistency within the whole image) and local attention (fined-grained details). This approach results in faster generation speed, higher generation fidelity, and improved resolution. (3) We propose a new generation pipeline incorporating autoencoding training and autoregressive generation strategy, demonstrating a better paradigm for image synthesis. Extensive experiments demonstrate the superiority of our approach in high-quality and high-resolution image reconstruction and generation.

Yueqin Yin, Lianghua Huang, Yu Liu et al.

Recent generative models show impressive results in photo-realistic image generation. However, artifacts often inevitably appear in the generated results, leading to downgraded user experience and reduced performance in downstream tasks. This work aims to develop a plugin post-processing module for diverse generative models, which can faithfully restore images from diverse generative artifacts. This is challenging because: (1) Unlike traditional degradation patterns, generative artifacts are non-linear and the transformation function is highly complex. (2) There are no readily available artifact-image pairs. (3) Different from model-specific anti-artifact methods, a model-agnostic framework views the generator as a black-box machine and has no access to the architecture details. In this work, we first design a group of mechanisms to simulate generative artifacts of popular generators (i.e., GANs, autoregressive models, and diffusion models), given real images. Second, we implement the model-agnostic anti-artifact framework as an image-to-image diffusion model, due to its advantage in generation quality and capacity. Finally, we design a conditioning scheme for the diffusion model to enable both blind and non-blind image restoration. A guidance parameter is also introduced to allow for a trade-off between restoration accuracy and image quality. Extensive experiments show that our method significantly outperforms previous approaches on the proposed datasets and real-world artifact images.

Jinbo Xing, Zeyinzi Jiang, Yuxiang Tuo et al.

Recent unified models have made unprecedented progress in both understanding and generation. However, while most of them accept multi-modal inputs, they typically produce only single-modality outputs. This challenge of producing interleaved content is mainly due to training data scarcity and the difficulty of modeling long-range cross-modal context. To address this issue, we decompose interleaved generation into textual planning and visual consistency modeling, and introduce a framework consisting of a planner and a visualizer. The planner produces dense textual descriptions for visual content, while the visualizer synthesizes images accordingly. Under this guidance, we construct large-scale textual-proxy interleaved data (where visual content is represented in text) to train the planner, and curate reference-guided image data to train the visualizer. These designs give rise to Wan-Weaver, which exhibits emergent interleaved generation ability with long-range textual coherence and visual consistency. Meanwhile, the integration of diverse understanding and generation data into planner training enables Wan-Weaver to achieve robust task reasoning and generation proficiency. To assess the model's capability in interleaved generation, we further construct a benchmark that spans a wide range of use cases across multiple dimensions. Extensive experiments demonstrate that, even without access to any real interleaved data, Wan-Weaver achieves superior performance over existing methods.

Lianghua Huang, Xin Zhao, Kaiqi Huang

A key capability of a long-term tracker is to search for targets in very large areas (typically the entire image) to handle possible target absences or tracking failures. However, currently there is a lack of such a strong baseline for global instance search. In this work, we aim to bridge this gap. Specifically, we propose GlobalTrack, a pure global instance search based tracker that makes no assumption on the temporal consistency of the target's positions and scales. GlobalTrack is developed based on two-stage object detectors, and it is able to perform full-image and multi-scale search of arbitrary instances with only a single query as the guide. We further propose a cross-query loss to improve the robustness of our approach against distractors. With no online learning, no punishment on position or scale changes, no scale smoothing and no trajectory refinement, our pure global instance search based tracker achieves comparable, sometimes much better performance on four large-scale tracking benchmarks (i.e., 52.1% AUC on LaSOT, 63.8% success rate on TLP, 60.3% MaxGM on OxUvA and 75.4% normalized precision on TrackingNet), compared to state-of-the-art approaches that typically require complex post-processing. More importantly, our tracker runs without cumulative errors, i.e., any type of temporary tracking failures will not affect its performance on future frames, making it ideal for long-term tracking. We hope this work will be a strong baseline for long-term tracking and will stimulate future works in this area. Code is available at https://github.com/huanglianghua/GlobalTrack.

Shilong Zhang, Lianghua Huang, Xi Chen et al.

This work presents FlashFace, a practical tool with which users can easily personalize their own photos on the fly by providing one or a few reference face images and a text prompt. Our approach is distinguishable from existing human photo customization methods by higher-fidelity identity preservation and better instruction following, benefiting from two subtle designs. First, we encode the face identity into a series of feature maps instead of one image token as in prior arts, allowing the model to retain more details of the reference faces (e.g., scars, tattoos, and face shape ). Second, we introduce a disentangled integration strategy to balance the text and image guidance during the text-to-image generation process, alleviating the conflict between the reference faces and the text prompts (e.g., personalizing an adult into a "child" or an "elder"). Extensive experimental results demonstrate the effectiveness of our method on various applications, including human image personalization, face swapping under language prompts, making virtual characters into real people, etc. Project Page: https://jshilong.github.io/flashface-page.

Shenggan Cheng, Yuanxin Wei, Lansong Diao et al.

Leveraging the diffusion transformer (DiT) architecture, models like Sora, CogVideoX and Wan have achieved remarkable progress in text-to-video, image-to-video, and video editing tasks. Despite these advances, diffusion-based video generation remains computationally intensive, especially for high-resolution, long-duration videos. Prior work accelerates its inference by skipping computation, usually at the cost of severe quality degradation. In this paper, we propose SRDiffusion, a novel framework that leverages collaboration between large and small models to reduce inference cost. The large model handles high-noise steps to ensure semantic and motion fidelity (Sketching), while the smaller model refines visual details in low-noise steps (Rendering). Experimental results demonstrate that our method outperforms existing approaches, over 3$\times$ speedup for Wan with nearly no quality loss for VBench, and 2$\times$ speedup for CogVideoX. Our method is introduced as a new direction orthogonal to existing acceleration strategies, offering a practical solution for scalable video generation.

Lianghua Huang, Yu Liu, Bin Wang et al.

A key challenge in self-supervised video representation learning is how to effectively capture motion information besides context bias. While most existing works implicitly achieve this with video-specific pretext tasks (e.g., predicting clip orders, time arrows, and paces), we develop a method that explicitly decouples motion supervision from context bias through a carefully designed pretext task. Specifically, we take the keyframes and motion vectors in compressed videos (e.g., in H.264 format) as the supervision sources for context and motion, respectively, which can be efficiently extracted at over 500 fps on the CPU. Then we design two pretext tasks that are jointly optimized: a context matching task where a pairwise contrastive loss is cast between video clip and keyframe features; and a motion prediction task where clip features, passed through an encoder-decoder network, are used to estimate motion features in a near future. These two tasks use a shared video backbone and separate MLP heads. Experiments show that our approach improves the quality of the learned video representation over previous works, where we obtain absolute gains of 16.0% and 11.1% in video retrieval recall on UCF101 and HMDB51, respectively. Moreover, we find the motion prediction to be a strong regularization for video networks, where using it as an auxiliary task improves the accuracy of action recognition with a margin of 7.4%~13.8%.

Yu Liu, Lianghua Huang, Pan Pan et al.

This paper presents a simple unsupervised visual representation learning method with a pretext task of discriminating all images in a dataset using a parametric, instance-level classifier. The overall framework is a replica of a supervised classification model, where semantic classes (e.g., dog, bird, and ship) are replaced by instance IDs. However, scaling up the classification task from thousands of semantic labels to millions of instance labels brings specific challenges including 1) the large-scale softmax computation; 2) the slow convergence due to the infrequent visiting of instance samples; and 3) the massive number of negative classes that can be noisy. This work presents several novel techniques to handle these difficulties. First, we introduce a hybrid parallel training framework to make large-scale training feasible. Second, we present a raw-feature initialization mechanism for classification weights, which we assume offers a contrastive prior for instance discrimination and can clearly speed up converge in our experiments. Finally, we propose to smooth the labels of a few hardest classes to avoid optimizing over very similar negative pairs. While being conceptually simple, our framework achieves competitive or superior performance compared to state-of-the-art unsupervised approaches, i.e., SimCLR, MoCoV2, and PIC under ImageNet linear evaluation protocol and on several downstream visual tasks, verifying that full instance classification is a strong pretraining technique for many semantic visual tasks.

Lianghua Huang, Xin Zhao, Kaiqi Huang

We introduce here a large tracking database that offers an unprecedentedly wide coverage of common moving objects in the wild, called GOT-10k. Specifically, GOT-10k is built upon the backbone of WordNet structure and it populates the majority of over 560 classes of moving objects and 87 motion patterns, magnitudes wider than the most recent similar-scale counterparts. The contributions of this paper are summarized in the following: (1) GOT-10k offers over 10,000 video segments with more than 1.5 million manually labeled bounding boxes, enabling unified training and stable evaluation of deep trackers. (2) GOT-10k is by far the first video trajectory dataset that uses the semantic hierarchy of WordNet to guide class population. (3) For the first time, GOT-10k introduces the one-shot protocol for tracker evaluation, where the training and test classes are zero-overlapped. The protocol avoids biased evaluation results towards familiar objects and it promotes generalization in tracker development. (4) We conduct extensive tracking experiments with 39 typical tracking algorithms on GOT-10k and analyze their results in this paper. (5) Finally, we develop a comprehensive platform for the tracking community that offers full-featured evaluation toolkits, an online evaluation server, and a responsive leaderboard. The annotations of GOT-10k's test data are kept private to avoid tuning parameters on it. The database, toolkits, evaluation server and baseline results are available at http://got-10k.aitestunion.com.