Qi Dai

Rui Tian, Qi Dai, Jianmin Bao et al.

Commercial video generation models have exhibited realistic, high-fidelity results but are still restricted to limited access. One crucial obstacle for large-scale applications is the expensive training and inference cost. In this paper, we argue that videos contain significantly more redundant information than images, allowing them to be encoded with very few motion latents. Towards this goal, we design an image-conditioned VAE that projects videos into extremely compressed latent space and decode them based on content images. This magic Reducio charm enables 64x reduction of latents compared to a common 2D VAE, without sacrificing the quality. Building upon Reducio-VAE, we can train diffusion models for high-resolution video generation efficiently. Specifically, we adopt a two-stage generation paradigm, first generating a condition image via text-to-image generation, followed by text-image-to-video generation with the proposed Reducio-DiT. Extensive experiments show that our model achieves strong performance in evaluation. More importantly, our method significantly boosts the training and inference efficiency of video LDMs. Reducio-DiT is trained in just 3.2K A100 GPU hours in total and can generate a 16-frame 1024$\times$1024 video clip within 15.5 seconds on a single A100 GPU. Code released at https://github.com/microsoft/Reducio-VAE .

Zhenda Xie, Zheng Zhang, Yue Cao et al.

This paper presents SimMIM, a simple framework for masked image modeling. We simplify recently proposed related approaches without special designs such as block-wise masking and tokenization via discrete VAE or clustering. To study what let the masked image modeling task learn good representations, we systematically study the major components in our framework, and find that simple designs of each component have revealed very strong representation learning performance: 1) random masking of the input image with a moderately large masked patch size (e.g., 32) makes a strong pre-text task; 2) predicting raw pixels of RGB values by direct regression performs no worse than the patch classification approaches with complex designs; 3) the prediction head can be as light as a linear layer, with no worse performance than heavier ones. Using ViT-B, our approach achieves 83.8% top-1 fine-tuning accuracy on ImageNet-1K by pre-training also on this dataset, surpassing previous best approach by +0.6%. When applied on a larger model of about 650 million parameters, SwinV2-H, it achieves 87.1% top-1 accuracy on ImageNet-1K using only ImageNet-1K data. We also leverage this approach to facilitate the training of a 3B model (SwinV2-G), that by $40\times$ less data than that in previous practice, we achieve the state-of-the-art on four representative vision benchmarks. The code and models will be publicly available at https://github.com/microsoft/SimMIM.

Baifeng Shi, Dinghuai Zhang, Qi Dai et al.

Convolutional Neural Networks (CNNs) are known to rely more on local texture rather than global shape when making decisions. Recent work also indicates a close relationship between CNN's texture-bias and its robustness against distribution shift, adversarial perturbation, random corruption, etc. In this work, we attempt at improving various kinds of robustness universally by alleviating CNN's texture bias. With inspiration from the human visual system, we propose a light-weight model-agnostic method, namely Informative Dropout (InfoDrop), to improve interpretability and reduce texture bias. Specifically, we discriminate texture from shape based on local self-information in an image, and adopt a Dropout-like algorithm to decorrelate the model output from the local texture. Through extensive experiments, we observe enhanced robustness under various scenarios (domain generalization, few-shot classification, image corruption, and adversarial perturbation). To the best of our knowledge, this work is one of the earliest attempts to improve different kinds of robustness in a unified model, shedding new light on the relationship between shape-bias and robustness, also on new approaches to trustworthy machine learning algorithms. Code is available at https://github.com/bfshi/InfoDrop.

Yupan Huang, Qi Dai, Yutong Lu

Video temporal action detection aims to temporally localize and recognize the action in untrimmed videos. Existing one-stage approaches mostly focus on unifying two subtasks, i.e., localization of action proposals and classification of each proposal through a fully shared backbone. However, such design of encapsulating all components of two subtasks in one single network might restrict the training by ignoring the specialized characteristic of each subtask. In this paper, we propose a novel Decoupled Single Shot temporal Action Detection (Decouple-SSAD) method to mitigate such problem by decoupling the localization and classification in a one-stage scheme. Particularly, two separate branches are designed in parallel to enable each component to own representations privately for accurate localization or classification. Each branch produces a set of action anchor layers by applying deconvolution to the feature maps of the main stream. Each branch produces a set of feature maps by applying deconvolution to the feature maps of the main stream. High-level semantic information from deeper layers is thus incorporated to enhance the feature representations. We conduct extensive experiments on THUMOS14 dataset and demonstrate superior performance over state-of-the-art methods. Our code is available online.

Shuyuan Tu, Zhen Xing, Xintong Han et al.

Current diffusion models for human image animation struggle to ensure identity (ID) consistency. This paper presents StableAnimator, the first end-to-end ID-preserving video diffusion framework, which synthesizes high-quality videos without any post-processing, conditioned on a reference image and a sequence of poses. Building upon a video diffusion model, StableAnimator contains carefully designed modules for both training and inference striving for identity consistency. In particular, StableAnimator begins by computing image and face embeddings with off-the-shelf extractors, respectively and face embeddings are further refined by interacting with image embeddings using a global content-aware Face Encoder. Then, StableAnimator introduces a novel distribution-aware ID Adapter that prevents interference caused by temporal layers while preserving ID via alignment. During inference, we propose a novel Hamilton-Jacobi-Bellman (HJB) equation-based optimization to further enhance the face quality. We demonstrate that solving the HJB equation can be integrated into the diffusion denoising process, and the resulting solution constrains the denoising path and thus benefits ID preservation. Experiments on multiple benchmarks show the effectiveness of StableAnimator both qualitatively and quantitatively.

Weiquan Huang, Aoqi Wu, Yifan Yang et al.

CLIP is a foundational multimodal model that aligns image and text features into a shared representation space via contrastive learning on large-scale image-text pairs. Its effectiveness primarily stems from the use of natural language as rich supervision. Motivated by the remarkable advancements in large language models (LLMs), this work explores how LLMs' superior text understanding and extensive open-world knowledge can enhance CLIP's capability, especially for processing longer and more complex image captions. We propose an efficient post-training strategy that integrates LLMs into pretrained CLIP. To address the challenge posed by the autoregressive nature of LLMs, we introduce a caption-to-caption contrastive fine-tuning framework, significantly enhancing the discriminative quality of LLM outputs. Extensive experiments demonstrate that our approach outperforms LoRA-based methods, achieving nearly fourfold faster training with superior performance. Furthermore, we validate substantial improvements over state-of-the-art models such as CLIP, EVA02, and SigLip2 across various zero-shot multimodal retrieval tasks, cross-lingual retrieval tasks, and multimodal language model pretraining.

Shuyuan Tu, Yueming Pan, Yinming Huang et al.

Current diffusion models for audio-driven avatar video generation struggle to synthesize long videos with natural audio synchronization and identity consistency. This paper presents StableAvatar, the first end-to-end video diffusion transformer that synthesizes infinite-length high-quality videos without post-processing. Conditioned on a reference image and audio, StableAvatar integrates tailored training and inference modules to enable infinite-length video generation. We observe that the main reason preventing existing models from generating long videos lies in their audio modeling. They typically rely on third-party off-the-shelf extractors to obtain audio embeddings, which are then directly injected into the diffusion model via cross-attention. Since current diffusion backbones lack any audio-related priors, this approach causes severe latent distribution error accumulation across video clips, leading the latent distribution of subsequent segments to drift away from the optimal distribution gradually. To address this, StableAvatar introduces a novel Time-step-aware Audio Adapter that prevents error accumulation via time-step-aware modulation. During inference, we propose a novel Audio Native Guidance Mechanism to further enhance the audio synchronization by leveraging the diffusion's own evolving joint audio-latent prediction as a dynamic guidance signal. To enhance the smoothness of the infinite-length videos, we introduce a Dynamic Weighted Sliding-window Strategy that fuses latent over time. Experiments on benchmarks show the effectiveness of StableAvatar both qualitatively and quantitatively.

Chendong Wang, Donglin Bai, Yifan Yang et al.

We present \emph{Video-in-the-Loop} (ViTL), a two-stage long-video QA framework that preserves a fixed token budget by first \emph{localizing} question-relevant interval(s) with a low-fps skim and then \emph{answering} via span-aware reallocation of visual tokens at higher effective frame rate, emitting an interleaved output with both spans and the final option for direct attribution. We also introduce \dataname{}, which converts description based event graphs into \emph{span-grounded} multiple-choice QA by pairing each question with \emph{ground-truth} time span(s) and related reasoning. ViTL is trained end-to-end with an interleaved group-relative objective that couples temporal IoU for localization with answer correctness, allowing credit to flow from answers back to spans without increasing compute. Under fixed token budgets, ViTL attains up to 8.6% with 50% less frame input on long-video QA and temporal grounding (e.g., Charades-STA, ActivityNet-Captions) and ablations show that span-aware token reallocation consistently surpasses uniform sampling. Together, \dataname{} and ViTL provide an interpretable, compute-efficient recipe for scalable long-video QA.

Jisheng Li, Qi Dai, Jiangtao Wen

Consistent in-focus input imagery is an essential precondition for machine vision systems to perceive the dynamic environment. A defocus blur severely degrades the performance of vision systems. To tackle this problem, we propose a deep-learning-based framework estimating the kernel scale and orientation of the defocus blur to adjust lens focus rapidly. Our pipeline utilizes 3D ConvNet for a variable number of input hypotheses to select the optimal slice from the input stack. We use random shuffle and Gumbel-softmax to improve network performance. We also propose to generate synthetic defocused images with various asymmetric coded apertures to facilitate training. Experiments are conducted to demonstrate the effectiveness of our framework.

Baifeng Shi, Qi Dai, Judy Hoffman et al.

Training temporal action detection in videos requires large amounts of labeled data, yet such annotation is expensive to collect. Incorporating unlabeled or weakly-labeled data to train action detection model could help reduce annotation cost. In this work, we first introduce the Semi-supervised Action Detection (SSAD) task with a mixture of labeled and unlabeled data and analyze different types of errors in the proposed SSAD baselines which are directly adapted from the semi-supervised classification task. To alleviate the main error of action incompleteness (i.e., missing parts of actions) in SSAD baselines, we further design an unsupervised foreground attention (UFA) module utilizing the "independence" between foreground and background motion. Then we incorporate weakly-labeled data into SSAD and propose Omni-supervised Action Detection (OSAD) with three levels of supervision. An information bottleneck (IB) suppressing the scene information in non-action frames while preserving the action information is designed to help overcome the accompanying action-context confusion problem in OSAD baselines. We extensively benchmark against the baselines for SSAD and OSAD on our created data splits in THUMOS14 and ActivityNet1.2, and demonstrate the effectiveness of the proposed UFA and IB methods. Lastly, the benefit of our full OSAD-IB model under limited annotation budgets is shown by exploring the optimal annotation strategy for labeled, unlabeled and weakly-labeled data.

Xingbo Liu, Xiushan Nie, Qi Dai et al.

Due to the compelling efficiency in retrieval and storage, similarity-preserving hashing has been widely applied to approximate nearest neighbor search in large-scale image retrieval. However, existing methods have poor performance in retrieval using an extremely short-length hash code due to weak ability of classification and poor distribution of hash bit. To address this issue, in this study, we propose a novel reinforcing short-length hashing (RSLH). In this proposed RSLH, mutual reconstruction between the hash representation and semantic labels is performed to preserve the semantic information. Furthermore, to enhance the accuracy of hash representation, a pairwise similarity matrix is designed to make a balance between accuracy and training expenditure on memory. In addition, a parameter boosting strategy is integrated to reinforce the precision with hash bits fusion. Extensive experiments on three large-scale image benchmarks demonstrate the superior performance of RSLH under various short-length hashing scenarios.

Baifeng Shi, Qi Dai, Yadong Mu et al.

Weakly-supervised temporal action localization is a problem of learning an action localization model with only video-level action labeling available. The general framework largely relies on the classification activation, which employs an attention model to identify the action-related frames and then categorizes them into different classes. Such method results in the action-context confusion issue: context frames near action clips tend to be recognized as action frames themselves, since they are closely related to the specific classes. To solve the problem, in this paper we propose to model the class-agnostic frame-wise probability conditioned on the frame attention using conditional Variational Auto-Encoder (VAE). With the observation that the context exhibits notable difference from the action at representation level, a probabilistic model, i.e., conditional VAE, is learned to model the likelihood of each frame given the attention. By maximizing the conditional probability with respect to the attention, the action and non-action frames are well separated. Experiments on THUMOS14 and ActivityNet1.2 demonstrate advantage of our method and effectiveness in handling action-context confusion problem. Code is now available on GitHub.

Qi Dai, Vaishakh Patil, Simon Hecker et al.

We present a self-supervised learning framework to estimate the individual object motion and monocular depth from video. We model the object motion as a 6 degree-of-freedom rigid-body transformation. The instance segmentation mask is leveraged to introduce the information of object. Compared with methods which predict dense optical flow map to model the motion, our approach significantly reduces the number of values to be estimated. Our system eliminates the scale ambiguity of motion prediction through imposing a novel geometric constraint loss term. Experiments on KITTI driving dataset demonstrate our system is capable to capture the object motion without external annotation. Our system outperforms previous self-supervised approaches in terms of 3D scene flow prediction, and contribute to the disparity prediction in dynamic area.

Zhi-Qi Cheng, Jun-Xiu Li, Qi Dai et al.

Tremendous variation in the scale of people/head size is a critical problem for crowd counting. To improve the scale invariance of feature representation, recent works extensively employ Convolutional Neural Networks with multi-column structures to handle different scales and resolutions. However, due to the substantial redundant parameters in columns, existing multi-column networks invariably exhibit almost the same scale features in different columns, which severely affects counting accuracy and leads to overfitting. In this paper, we attack this problem by proposing a novel Multi-column Mutual Learning (McML) strategy. It has two main innovations: 1) A statistical network is incorporated into the multi-column framework to estimate the mutual information between columns, which can approximately indicate the scale correlation between features from different columns. By minimizing the mutual information, each column is guided to learn features with different image scales. 2) We devise a mutual learning scheme that can alternately optimize each column while keeping the other columns fixed on each mini-batch training data. With such asynchronous parameter update process, each column is inclined to learn different feature representation from others, which can efficiently reduce the parameter redundancy and improve generalization ability. More remarkably, McML can be applied to all existing multi-column networks and is end-to-end trainable. Extensive experiments on four challenging benchmarks show that McML can significantly improve the original multi-column networks and outperform the other state-of-the-art approaches.

Zhi-Qi Cheng, Jun-Xiu Li, Qi Dai et al.

The aim of crowd counting is to estimate the number of people in images by leveraging the annotation of center positions for pedestrians' heads. Promising progresses have been made with the prevalence of deep Convolutional Neural Networks. Existing methods widely employ the Euclidean distance (i.e., $L_2$ loss) to optimize the model, which, however, has two main drawbacks: (1) the loss has difficulty in learning the spatial awareness (i.e., the position of head) since it struggles to retain the high-frequency variation in the density map, and (2) the loss is highly sensitive to various noises in crowd counting, such as the zero-mean noise, head size changes, and occlusions. Although the Maximum Excess over SubArrays (MESA) loss has been previously proposed to address the above issues by finding the rectangular subregion whose predicted density map has the maximum difference from the ground truth, it cannot be solved by gradient descent, thus can hardly be integrated into the deep learning framework. In this paper, we present a novel architecture called SPatial Awareness Network (SPANet) to incorporate spatial context for crowd counting. The Maximum Excess over Pixels (MEP) loss is proposed to achieve this by finding the pixel-level subregion with high discrepancy to the ground truth. To this end, we devise a weakly supervised learning scheme to generate such region with a multi-branch architecture. The proposed framework can be integrated into existing deep crowd counting methods and is end-to-end trainable. Extensive experiments on four challenging benchmarks show that our method can significantly improve the performance of baselines. More remarkably, our approach outperforms the state-of-the-art methods on all benchmark datasets.