As a fundamental backbone for video generation, diffusion models are challenged by low inference speed due to the sequential nature of denoising. Previous methods speed up the models by caching and reusing model outputs at uniformly selected timesteps. However, such a strategy neglects the fact that differences among model outputs are not uniform across timesteps, which hinders selecting the appropriate model outputs to cache, leading to a poor balance between inference efficiency and visual quality. In this study, we introduce Timestep Embedding Aware Cache (TeaCache), a training-free caching approach that estimates and leverages the fluctuating differences among model outputs across timesteps. Rather than directly using the time-consuming model outputs, TeaCache focuses on model inputs, which have a strong correlation with the modeloutputs while incurring negligible computational cost. TeaCache first modulates the noisy inputs using the timestep embeddings to ensure their differences better approximating those of model outputs. TeaCache then introduces a rescaling strategy to refine the estimated differences and utilizes them to indicate output caching. Experiments show that TeaCache achieves up to 4.41× acceleration over Open-Sora-Plan with negligible (-0.07% Vbench score) degradation of visual quality.

The current text-to-video (T2V) generation has made significant progress in synthesizing realistic general videos, but it is still under-explored in identity-specific human video generation with customized ID images. The key challenge lies in maintaining high ID fidelity consistently while preserving the original motion dynamic and semantic following after the identity injection. Current video identity customization methods mainly rely on reconstructing given identity images on text-to-image models, which have a divergent distribution with the T2V model. This process introduces a tuning-inference gap, leading to dynamic and semantic degradation. To tackle this problem, we propose a novel framework, dubbed $\textbf{PersonalVideo}$, that applies a mixture of reward supervision on synthesized videos instead of the simple reconstruction objective on images. Specifically, we first incorporate identity consistency reward to effectively inject the reference's identity without the tuning-inference gap. Then we propose a novel semantic consistency reward to align the semantic distribution of the generated videos with the original T2V model, which preserves its dynamic and semantic following capability during the identity injection. With the non-reconstructive reward training, we further employ simulated prompt augmentation to reduce overfitting by supervising generated results in more semantic scenarios, gaining good robustness even with only a single reference image. Extensive experiments demonstrate our method's superiority in delivering high identity faithfulness while preserving the inherent video generation qualities of the original T2V model, outshining prior methods.

Visual diffusion models achieve remarkable progress, yet they are typically trained at limited resolutions due to the lack of high-resolution data and constrained computation resources, hampering their ability to generate high-fidelity images or videos at higher resolutions. Recent efforts have explored tuning-free strategies to exhibit the untapped potential higher-resolution visual generation of pre-trained models. However, these methods are still prone to producing low-quality visual content with repetitive patterns. The key obstacle lies in the inevitable increase in high-frequency information when the model generates visual content exceeding its training resolution, leading to undesirable repetitive patterns deriving from the accumulated errors. To tackle this challenge, we propose FreeScale, a tuning-free inference paradigm to enable higher-resolution visual generation via scale fusion. Specifically, FreeScale processes information from different receptive scales and then fuses it by extracting desired frequency components. Extensive experiments validate the superiority of our paradigm in extending the capabilities of higher-resolution visual generation for both image and video models. Notably, compared with previous best-performing methods, FreeScale unlocks the 8k-resolution text-to-image generation for the first time.

With the availability of large-scale video datasets and the advances of diffusion models, text-driven video generation has achieved substantial progress. However, existing video generation models are typically trained on a limited number of frames, resulting in the inability to generate high-fidelity long videos during inference. Furthermore, these models only support single-text conditions, whereas real-life scenarios often require multi-text conditions as the video content changes over time. To tackle these challenges, this study explores the potential of extending the text-driven capability to generate longer videos conditioned on multiple texts. 1) We first analyze the impact of initial noise in video diffusion models. Then building upon the observation of noise, we propose FreeNoise, a tuning-free and time-efficient paradigm to enhance the generative capabilities of pretrained video diffusion models while preserving content consistency. Specifically, instead of initializing noises for all frames, we reschedule a sequence of noises for long-range correlation and perform temporal attention over them by window-based fusion. 2) Additionally, we design a novel motion injection method to support the generation of videos conditioned on multiple text prompts. Extensive experiments validate the superiority of our paradigm in extending the generative capabilities of video diffusion models. It is noteworthy that compared with the previous best-performing method which brought about 255% extra time cost, our method incurs only negligible time cost of approximately 17%. Generated video samples are available at our website: http://haonanqiu.com/projects/FreeNoise.html.

Generating high-quality and diverse human images is an important yet challenging task in vision and graphics. However, existing generative models often fall short under the high diversity of clothing shapes and textures. Furthermore, the generation process is even desired to be intuitively controllable for layman users. In this work, we present a text-driven controllable framework, Text2Human, for a high-quality and diverse human generation. We synthesize full-body human images starting from a given human pose with two dedicated steps. 1) With some texts describing the shapes of clothes, the given human pose is first translated to a human parsing map. 2) The final human image is then generated by providing the system with more attributes about the textures of clothes. Specifically, to model the diversity of clothing textures, we build a hierarchical texture-aware codebook that stores multi-scale neural representations for each type of texture. The codebook at the coarse level includes the structural representations of textures, while the codebook at the fine level focuses on the details of textures. To make use of the learned hierarchical codebook to synthesize desired images, a diffusion-based transformer sampler with mixture of experts is firstly employed to sample indices from the coarsest level of the codebook, which then is used to predict the indices of the codebook at finer levels. The predicted indices at different levels are translated to human images by the decoder learned accompanied with hierarchical codebooks. The use of mixture-of-experts allows for the generated image conditioned on the fine-grained text input. The prediction for finer level indices refines the quality of clothing textures. Extensive quantitative and qualitative evaluations demonstrate that our proposed Text2Human framework can generate more diverse and realistic human images compared to state-of-the-art methods. Our project page is https://yumingj.github.io/projects/Text2Human.html. Code and pretrained models are available at https://github.com/yumingj/Text2Human.

Recent studies show that convolutional neural networks (CNNs) are vulnerable under various settings, including adversarial attacks, common corruptions, and backdoor attacks. Motivated by the findings that human visual sys-tem pays more attention to global structure (e.g., shapes) for recognition while CNNs are biased towards local texture features in images, in this work we aim to analyze whether "edge features" could improve the recognition robustness in these scenarios, and if so, to what extent? To answer these questions and systematically evaluate the global structure features, we focus on shape features and pro-pose two edge-enabled pipelines EdgeNetRob and Edge-GANRob, forcing the CNNs to rely more on edge features. Specifically, EdgeNetRob and EdgeGANRob first explicitly extract shape structure features from a given image via an edge detection algorithm. Then EdgeNetRob trains down-stream learning tasks directly on the extracted edge features, while EdgeGANRob reconstructs a new image by refilling the texture information with a trained generative adversarial network (GANs). To reduce the sensitivity of edge detection algorithms to perturbations, we additionally propose a robust edge detection approach Robust Canny based on vanilla Canny. Based on our evaluation, we find that EdgeNetRob can help boost model robustness under different attack scenarios at the cost of the clean model accuracy. EdgeGANRob, on the other hand, is able to improve the clean model accuracy compared to EdgeNetRob while preserving the robustness. This shows that given such edge features, how to leverage them matters for robustness, and it also depends on data properties. Our systematic studies on edge structure features under different settings will shed light on future robust feature exploration and optimization.

The egocentric video provides a unique view of event participants to show their attention, vision, and interaction with objects. In this paper, we introduce Ego-Deliver, a new large-scale egocentric video benchmark recorded by takeaway riders about their daily work. To the best of our knowledge, Ego-Deliver presents the first attempt in understanding activities from the takeaway delivery process while being one of the largest egocentric video action datasets to date. Our dataset provides a total of 5,360 videos with more than 139,000 multi-track annotations and 45 different attributes, which we believe is pivotal to future research in this area. We introduce the FS-Net architecture, a new anchor-free action detection approach handling extreme variations of action durations. We partition videos into fragments and build dynamic graphs over fragments, where multi-fragment context information is aggregated to boost fragment classification. A splicing and scoring module is applied to obtain final action proposals. Our experimental evaluation confirms that the proposed framework outperforms existing approaches on the proposed Ego-Deliver benchmark and is competitive on other popular benchmarks. In our current version, Ego-Deliver is used to make a comprehensive comparison between algorithms for activity detection. We also show its application to action recognition with promising results. The dataset, toolkits and baseline results will be made available at: https://egodeliver.github.io/EgoDeliver_Dataset/

Video emotion recognition is a challenging task due to complex scenes and various forms of emotion expression. Most existing works focus on fusing multiple features over the whole video clips. According to our observations, given a long video clip, the emotion is usually presented by only several actions/objects in a few short snippets, and the meaningful cues are buried in the noisy background. When human judging the emotion in videos, we first find the informative clips and then closely look for emotional cues in the frames. In this paper, we propose Dual Focus Attention Network to mimic this process. First, three kinds of features including action, object, and scene are extracted from videos. Second, Two attention modules are used to focus on the visual features of the videos from temporal and spatial dimensions respectively. With our dual focus attention network, we can effectively discover the most emotional frames along the time dimension and the most emotional visual cues in each frame. Our experiments conducted on two widely used datasets Ekman and VideoEmotion show that our proposed approach outperforms the existing approaches.

Abstract Generating plausible hair image given limited guidance, such as sparse sketches or low‐resolution image, has been made possible with the rise of Generative Adversarial Networks (GANs). Traditional image‐to‐image translation networks can generate recognizable results, but finer textures are usually lost and blur artifacts commonly exist. In this paper, we propose a two‐phase generative model for high‐quality hair image synthesis. The two‐phase pipeline first generates a coarse image by an existing image translation model, then applies a re‐generating network with self‐enhancing capability to the coarse image. The self‐enhancing capability is achieved by a proposed differentiable layer, which extracts the structural texture and orientation maps from a hair image. Extensive experiments on two tasks, Sketch2Hair and Hair Super‐Resolution, demonstrate that our approach is able to synthesize plausible hair image with finer details, and reaches the state‐of‐the‐art.

Remote Direct Memory Access (RDMA) technique allows the messaging service that directly access the memory on remote machines, which provides low CPU overhead, low latency, and high throughput network transmission. On the other hand, however, due to the limited cache space in RDMA NIC (RNIC), it is still challenging to achieve effective and fair resource sharing across different applications. To address this problem, we present a scalable RDMA as a service to manage resource and deliver fair scheduling to applications' requests. We study the thread contention and preemptive schedule issues at end-hosts, and report the corresponding performance degradation through experiments. Then, we introduce Avatar, a model to manage memory and Queue Pairs (QPs) resource for a large number of connections, which eliminates the lock contention and provides fair data scheduling for applications with different priorities. Finally, we implement Avatar and demonstrate that Avatar can support a thousand of connections, improve the fairness and reduce the requests completion time up to 50% in comparison with the native RDMA.

Locating actions in long untrimmed videos has been a challenging problem in video content analysis. The performances of existing action localization approaches remain unsatisfactory in precisely determining the beginning and the end of an action. Imitating the human perception procedure with observations and refinements, we propose a novel three-phase action localization framework. Our framework is embedded with an Actionness Network to generate initial proposals through frame-wise similarity grouping, and then a Refinement Network to conduct boundary adjustment on these proposals. Finally, the refined proposals are sent to a Localization Network for further fine-grained location regression. The whole process can be deemed as multi-stage refinement using a novel non-local pyramid feature under various temporal granularities. We evaluate our framework on THUMOS14 benchmark and obtain a significant improvement over the state-of-the-arts approaches. Specifically, the performance gain is remarkable under precise localization with high IoU thresholds. Our proposed framework achieves [email protected]=0.5 of 34.2%.