Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model

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 4.41x acceleration over Open-Sora-Plan with negligible (-0.07% Vbench score) degradation of visual quality.

Evaluation of Text-to-Video Generation Models: A Dynamics Perspective

Comprehensive and constructive evaluation protocols play an important role in the development of sophisticated text-to-video (T2V) generation models. Existing evaluation protocols primarily focus on temporal consistency and content continuity, yet largely ignore the dynamics of video content. Dynamics are an essential dimension for measuring the visual vividness and the honesty of video content to text prompts. In this study, we propose an effective evaluation protocol, termed DEVIL, which centers on the dynamics dimension to evaluate T2V models. For this purpose, we establish a new benchmark comprising text prompts that fully reflect multiple dynamics grades, and define a set of dynamics scores corresponding to various temporal granularities to comprehensively evaluate the dynamics of each generated video. Based on the new benchmark and the dynamics scores, we assess T2V models with the design of three metrics: dynamics range, dynamics controllability, and dynamics-based quality. Experiments show that DEVIL achieves a Pearson correlation exceeding 90% with human ratings, demonstrating its potential to advance T2V generation models. Code is available at https://github.com/MingXiangL/DEVIL.

DynRefer: Delving into Region-level Multi-modality Tasks via Dynamic Resolution

Region-level multi-modality methods can translate referred image regions to human preferred language descriptions. Unfortunately, most of existing methods using fixed visual inputs remain lacking the resolution adaptability to find out precise language descriptions. In this study, we propose a dynamic resolution approach, referred to as DynRefer, to pursue high-accuracy region-level referring through mimicking the resolution adaptability of human visual cognition. DynRefer first implements stochastic vision-language alignment. It aligns desired language descriptions of multi-modality tasks with images of stochastic resolution, which are constructed by nesting a set of views around the referred region. DynRefer then implements dynamic multi-modality referring, which is realized by selecting views based on image and language priors. This allows the visual information used for referring to better match human preferences, thereby improving the representational adaptability of region-level multi-modality models. Extensive experiments show that DynRefer brings mutual improvement upon tasks including region-level captioning, open-vocabulary region recognition and attribute detection. Last but not least, DynRefer achieves new state-of-the-art on multiple region-level multi-modality tasks using a single model. Code is available at https://github.com/callsys/DynRefer.

Controllable Dense Captioner with Multimodal Embedding Bridging

In this paper, we propose a controllable dense captioner (ControlCap), which accommodates user's intention to dense captioning by introducing linguistic guidance. ControlCap is defined as a multimodal embedding bridging architecture, which comprises multimodal embedding generation (MEG) module and bi-directional embedding bridging (BEB) module. While MEG module represents objects/regions by combining embeddings of detailed information with context-aware ones, it also endows ControlCap the adaptability to specialized controls by utilizing them as linguistic guidance. BEB module aligns the linguistic guidance with visual embeddings through borrowing/returning features from/to the visual domain and gathering such features to predict text descriptions. Experiments on Visual Genome and VG-COCO datasets show that ControlCap respectively outperforms the state-of-the-art methods by 1.5% and 3.7% (mAP). Last but not least, with the capability of converting region-category pairs to region-text pairs, ControlCap is able to act as a powerful data engine for dense captioning. Code is available at https://github.com/callsys/ControlCap.

VMamba: Visual State Space Model

Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) stand as the two most popular foundation models for visual representation learning. While CNNs exhibit remarkable scalability with linear complexity w.r.t. image resolution, ViTs surpass them in fitting capabilities despite contending with quadratic complexity. A closer inspection reveals that ViTs achieve superior visual modeling performance through the incorporation of global receptive fields and dynamic weights. This observation motivates us to propose a novel architecture that inherits these components while enhancing computational efficiency. To this end, we draw inspiration from the recently introduced state space model and propose the Visual State Space Model (VMamba), which achieves linear complexity without sacrificing global receptive fields. To address the encountered direction-sensitive issue, we introduce the Cross-Scan Module (CSM) to traverse the spatial domain and convert any non-causal visual image into order patch sequences. Extensive experimental results substantiate that VMamba not only demonstrates promising capabilities across various visual perception tasks, but also exhibits more pronounced advantages over established benchmarks as the image resolution increases. Source code has been available at https://github.com/MzeroMiko/VMamba.

Continual Learning for Image Segmentation with Dynamic Query

Image segmentation based on continual learning exhibits a critical drop of performance, mainly due to catastrophic forgetting and background shift, as they are required to incorporate new classes continually. In this paper, we propose a simple, yet effective Continual Image Segmentation method with incremental Dynamic Query (CISDQ), which decouples the representation learning of both old and new knowledge with lightweight query embedding. CISDQ mainly includes three contributions: 1) We define dynamic queries with adaptive background class to exploit past knowledge and learn future classes naturally. 2) CISDQ proposes a class/instance-aware Query Guided Knowledge Distillation strategy to overcome catastrophic forgetting by capturing the inter-class diversity and intra-class identity. 3) Apart from semantic segmentation, CISDQ introduce the continual learning for instance segmentation in which instance-wise labeling and supervision are considered. Extensive experiments on three datasets for two tasks (i.e., continual semantic and instance segmentation are conducted to demonstrate that CISDQ achieves the state-of-the-art performance, specifically, obtaining 4.4% and 2.9% mIoU improvements for the ADE 100-10 (6 steps) setting and ADE 100-5 (11 steps) setting.

DatasetDM: Synthesizing Data with Perception Annotations Using Diffusion Models

Current deep networks are very data-hungry and benefit from training on largescale datasets, which are often time-consuming to collect and annotate. By contrast, synthetic data can be generated infinitely using generative models such as DALL-E and diffusion models, with minimal effort and cost. In this paper, we present DatasetDM, a generic dataset generation model that can produce diverse synthetic images and the corresponding high-quality perception annotations (e.g., segmentation masks, and depth). Our method builds upon the pre-trained diffusion model and extends text-guided image synthesis to perception data generation. We show that the rich latent code of the diffusion model can be effectively decoded as accurate perception annotations using a decoder module. Training the decoder only needs less than 1% (around 100 images) manually labeled images, enabling the generation of an infinitely large annotated dataset. Then these synthetic data can be used for training various perception models for downstream tasks. To showcase the power of the proposed approach, we generate datasets with rich dense pixel-wise labels for a wide range of downstream tasks, including semantic segmentation, instance segmentation, and depth estimation. Notably, it achieves 1) state-of-the-art results on semantic segmentation and instance segmentation; 2) significantly more robust on domain generalization than using the real data alone; and state-of-the-art results in zero-shot segmentation setting; and 3) flexibility for efficient application and novel task composition (e.g., image editing). The project website and code can be found at https://weijiawu.github.io/DatasetDM_page/ and https://github.com/showlab/DatasetDM, respectively

Generative Prompt Model for Weakly Supervised Object Localization

Weakly supervised object localization (WSOL) remains challenging when learning object localization models from image category labels. Conventional methods that discriminatively train activation models ignore representative yet less discriminative object parts. In this study, we propose a generative prompt model (GenPromp), defining the first generative pipeline to localize less discriminative object parts by formulating WSOL as a conditional image denoising procedure. During training, GenPromp converts image category labels to learnable prompt embeddings which are fed to a generative model to conditionally recover the input image with noise and learn representative embeddings. During inference, enPromp combines the representative embeddings with discriminative embeddings (queried from an off-the-shelf vision-language model) for both representative and discriminative capacity. The combined embeddings are finally used to generate multi-scale high-quality attention maps, which facilitate localizing full object extent. Experiments on CUB-200-2011 and ILSVRC show that GenPromp respectively outperforms the best discriminative models by 5.2% and 5.6% (Top-1 Loc), setting a solid baseline for WSOL with the generative model. Code is available at https://github.com/callsys/GenPromp.

A Large Cross-Modal Video Retrieval Dataset with Reading Comprehension

Most existing cross-modal language-to-video retrieval (VR) research focuses on single-modal input from video, i.e., visual representation, while the text is omnipresent in human environments and frequently critical to understand video. To study how to retrieve video with both modal inputs, i.e., visual and text semantic representations, we first introduce a large-scale and cross-modal Video Retrieval dataset with text reading comprehension, TextVR, which contains 42.2k sentence queries for 10.5k videos of 8 scenario domains, i.e., Street View (indoor), Street View (outdoor), Games, Sports, Driving, Activity, TV Show, and Cooking. The proposed TextVR requires one unified cross-modal model to recognize and comprehend texts, relate them to the visual context, and decide what text semantic information is vital for the video retrieval task. Besides, we present a detailed analysis of TextVR compared to the existing datasets and design a novel multimodal video retrieval baseline for the text-based video retrieval task. The dataset analysis and extensive experiments show that our TextVR benchmark provides many new technical challenges and insights from previous datasets for the video-and-language community. The project website and GitHub repo can be found at https://sites.google.com/view/loveucvpr23/guest-track and https://github.com/callsys/TextVR, respectively.

FlowText: Synthesizing Realistic Scene Text Video with Optical Flow Estimation

Current video text spotting methods can achieve preferable performance, powered with sufficient labeled training data. However, labeling data manually is time-consuming and labor-intensive. To overcome this, using low-cost synthetic data is a promising alternative. This paper introduces a novel video text synthesis technique called FlowText, which utilizes optical flow estimation to synthesize a large amount of text video data at a low cost for training robust video text spotters. Unlike existing methods that focus on image-level synthesis, FlowText concentrates on synthesizing temporal information of text instances across consecutive frames using optical flow. This temporal information is crucial for accurately tracking and spotting text in video sequences, including text movement, distortion, appearance, disappearance, shelter, and blur. Experiments show that combining general detectors like TransDETR with the proposed FlowText produces remarkable results on various datasets, such as ICDAR2015video and ICDAR2013video. Code is available at https://github.com/callsys/FlowText.