Video Domain Research Articles

Spatio-Temporal Attention for Text-Video Retrieval

Text-video retrieval, a fundamental task for associating textual descriptions with video content, has become increasingly important in the video domain. Most existing methods focus on the single-modality features only considering the knowledge within individual video or text modalities, often neglecting cross-modal interactions. However, a text description corresponds to a specific spatio-temporal content within a video, involving a certain segment of a frame sequence and distinct sub-regions within these frames. Therefore, we focus on the text-conditioned video features to bridge the modality gap. In this paper, we propose Spatio-Temporal Attention for video-text retrieval, termed STAttn, which utilizes textual information to focus on the spatio-temporal video content. Our final text-conditioned video features are generated from the text-related video frames and the text-related regions within these frames. First, we propose the Spatial Text-Attention Module (STAM) to learn the spatial information within video frames. STAM introduces the text-related salient patches to capture more fine-grained details. Second, we propose the Temporal Text-Attention Module (TTAM) to learn the temporal relationships between video frames. Temporal Triplet loss is proposed in TTAM to enhance the attention towards the text-related frames. Thus, the two modules learn the text-related spatio-temporal content from both intra-frame and inter-frame aspects. Extensive experiments on three benchmark datasets, MSRVTT, ActivityNet, and DiDeMo, demonstrate that our STAttn outperforms state-of-the-art methods.

Toward Domain Adaptive Learning-Based Variation Autoencoder Emotional Analysis in English Teaching

Speech emotion analysis plays an important role in English teaching by analyzing the reading state of students. Teachers can dynamically adjust the teaching content according to the emotional feedback of students and improve the teaching quality of the school. Due to unstable student emotions and background noise, the accuracy of speech emotion recognition is constrained. Although multimodal data can alleviate the deficiency of a single modality, collecting and annotating multimodal samples requires a significant amount of resources. To resolve this issue, this paper proposes a novel multimodal sentiment analysis framework based on domain adaptive learning mechanisms to assist English teaching. We construct a novel multi-task variation autoencoder framework in which we simultaneously complete reconstruction and classification tasks. To improve speech emotion recognition performance, we introduce domain adaptive learning based on the Wasserstein distance between two variational hidden layers from the video domain (source domain) and speech domain (target domain). To validate the effectiveness of our proposed model, we conducted extensive comparative experiments on two public datasets and a self-built English oral dataset. All experimental results indicate that domain adaptation learning mechanisms can effectively improve the recognition performance of the target domain. On the self-built dataset for English teaching, the proposed model achieves higher performance compared to other deep models.

Percept, Chat, Adapt: Knowledge transfer of foundation models for open-world video recognition

Percept, Chat, Adapt: Knowledge transfer of foundation models for open-world video recognition

A Survey on Video Diffusion Models

The recent wave of AI-generated content (AIGC) has witnessed substantial success in computer vision, with the diffusion model playing a crucial role in this achievement. Due to their impressive generative capabilities, diffusion models are gradually superseding methods based on GANs and auto-regressive Transformers, demonstrating exceptional performance not only in image generation and editing, but also in the realm of video-related research. However, existing surveys mainly focus on diffusion models in the context of image generation, with few up-to-date reviews on their application in the video domain. To address this gap, this article presents a comprehensive review of video diffusion models in the AIGC era. Specifically, we begin with a concise introduction to the fundamentals and evolution of diffusion models. Subsequently, we present an overview of research on diffusion models in the video domain, categorizing the work into three key areas: video generation, video editing, and other video understanding tasks. We conduct a thorough review of the literature in these three key areas, including further categorization and practical contributions in the field. Finally, we discuss the challenges faced by research in this domain and outline potential future developmental trends. A comprehensive list of video diffusion models studied in this survey is available at https://github.com/ChenHsing/Awesome-Video-Diffusion-Models .

Retargeting Video With an End-to-End Framework.

Video holds significance in computer graphics applications. Because of the heterogeneous of digital devices, retargeting videos becomes an essential function to enhance user viewing experience in such applications. In the research of video retargeting, preserving the relevant visual content in videos, avoiding flicking, and processing time are the vital challenges. Extending image retargeting techniques to the video domain is challenging due to the high running time. Prior work of video retargeting mainly utilizes time-consuming preprocessing to analyze frames. Plus, being tolerant of different video content, avoiding important objects from shrinking, and the ability to play with arbitrary ratios are the limitations that need to be resolved in these systems requiring investigation. In this paper, we present an end-to-end RETVI method to retarget videos to arbitrary aspect ratios. We eliminate the computational bottleneck in the conventional approaches by designing RETVI with two modules, content feature analyzer (CFA) and adaptive deforming estimator (ADE). The extensive experiments and evaluations show that our system outperforms previous work in quality and running time.

Adapting CLIP for Action Recognition via Dual Semantic Supervision and Temporal Prompt Reparameterization

The contrastive vision–language pre-trained model CLIP, driven by large-scale open-vocabulary image–text pairs, has recently demonstrated remarkable zero-shot generalization capabilities in diverse downstream image tasks, which has made numerous models dominated by the “image pre-training followed by fine-tuning” paradigm exhibit promising results on standard video benchmarks. However, as models scale up, full fine-tuning adaptive strategy for specific tasks becomes difficult in terms of training and storage. In this work, we propose a novel method that adapts CLIP to the video domain for efficient recognition without destroying the original pre-trained parameters. Specifically, we introduce temporal prompts to realize the object of reasoning about the dynamic content of videos for pre-trained models that lack temporal cues. Then, by replacing the direct learning style of prompt vectors with a lightweight reparameterization encoder, the model can be adapted to domain-specific adjustment to learn more generalizable representations. Furthermore, we predefine a Chinese label dictionary to enhance video representation by co-supervision of Chinese and English semantics. Extensive experiments on video action recognition benchmarks show that our method achieves competitive or even better performance than most existing methods with fewer trainable parameters in both general and few-shot recognition scenarios.

Discrete Space Deep Reinforcement Learning Algorithm Based on Support Vector Machine Recursive Feature Elimination

Algorithms for training agents with experience replay have advanced in several domains, primarily because prioritized experience replay (PER) developed from the double deep Q-network (DDQN) in deep reinforcement learning (DRL) has become a standard. PER-based algorithms have achieved significant success in the image and video domains. However, the exceptional results observed in images and videos are not as effective in many domains with simple action spaces and relatively small states, particularly in discrete action spaces with sparse rewards. Moreover, most advanced techniques may improve sampling efficiency using deep learning algorithms rather than reinforcement learning. However, there is growing evidence that deep learning algorithms cannot generalize during training. Therefore, this study proposes an algorithm suitable for discrete action space environments that uses the sample efficiency of PER based on DDQN but incorporates support vector machine recursive feature elimination (SVM-RFE) without enhancing the sampling efficiency through deep learning algorithms. The proposed algorithm exhibited considerable performance improvements in classical OpenAI Gym environments that did not use images or videos as inputs. In particular, simple discrete space environments with reflection symmetry, such as Cart–Pole, exhibited a faster and more stable learning process. These results suggest that the application of SVM-RFE, which leverages the orthogonality of support vector machines (SVMs) across learning patterns, can be appropriate when the data in the reinforcement learning environment demonstrate symmetry.

Video Unsupervised Domain Adaptation with Deep Learning: A Comprehensive Survey

Video analysis tasks such as action recognition have received increasing research interest with growing applications in fields such as smart healthcare, thanks to the introduction of large-scale datasets and deep learning-based representations. However, video models trained on existing datasets suffer from significant performance degradation when deployed directly to real-world applications due to domain shifts between the training public video datasets (source video domains) and real-world videos (target video domains). Further, with the high cost of video annotation, it is more practical to use unlabeled videos for training. To tackle performance degradation and address concerns in high video annotation cost uniformly, the video unsupervised domain adaptation (VUDA) is introduced to adapt video models from the labeled source domain to the unlabeled target domain by alleviating video domain shift, improving the generalizability and portability of video models. This paper surveys recent progress in VUDA with deep learning. We begin with the motivation of VUDA, followed by its definition, and recent progress of methods for both closed-set VUDA and VUDA under different scenarios, and current benchmark datasets for VUDA research. Eventually, future directions are provided to promote further VUDA research. The repository of this survey is provided at https://github.com/xuyu0010/awesome-video-domain-adaptation.

Video domain adaptation for semantic segmentation using perceptual consistency matching

Unsupervised domain adaptation (UDA) aims to transfer knowledge in previous and related labeled datasets (sources) to a new unlabeled dataset (target). Despite the impressive performance, existing approaches have largely focused on image-based UDA only, and video-based UDA has been relatively understudied and received less attention due to the difficulty of adapting diverse modal video features and modeling temporal associations efficiently. To address this, existing studies use optical flow to capture motion cues between in-domain consecutive frames, but is limited by heavy compute requirements and modeling flow patterns across diverse domains is equally challenging. In this work, we propose an adversarial domain adaptation approach for video semantic segmentation that aims to align temporally associated pixels in successive source and target domain frames without relying on optical flow. Specifically, we introduce a Perceptual Consistency Matching (PCM) strategy that leverages perceptual similarity to identify pixels with high correlation across consecutive frames, and infer that such pixels should correspond to the same class. Therefore, we can enhance prediction accuracy for video-UDA by enforcing consistency not only between in-domain frames, but across domains using PCM objectives during model training. Extensive experiments on public datasets show the benefit of our approach over existing state-of-the-art UDA methods. Our approach not only addresses a crucial task in video domain adaptation but also offers notable improvements in performance with faster inference times.

Building an Open-Vocabulary Video CLIP Model With Better Architectures, Optimization and Data.

Despite significant results achieved by Contrastive Language-Image Pretraining (CLIP) in zero-shot image recognition, limited effort has been made exploring its potential for zero-shot video recognition. This paper presents Open-VCLIP++, a simple yet effective framework that adapts CLIP to a strong zero-shot video classifier, capable of identifying novel actions and events during testing. Open-VCLIP++ minimally modifies CLIP to capture spatial-temporal relationships in videos, thereby creating a specialized video classifier while striving for generalization. We formally demonstrate that training Open-VCLIP++ is tantamount to continual learning with zero historical data. To address this problem, we introduce Interpolated Weight Optimization, a technique that leverages the advantages of weight interpolation during both training and testing. Furthermore, we build upon large language models to produce fine-grained video descriptions. These detailed descriptions are further aligned with video features, facilitating a better transfer of CLIP to the video domain. Our approach is evaluated on three widely used action recognition datasets, following a variety of zero-shot evaluation protocols. The results demonstrate that our method surpasses existing state-of-the-art techniques by significant margins. Specifically, we achieve zero-shot accuracy scores of 88.1%, 58.7%, and 81.2% on UCF, HMDB, and Kinetics-600 datasets respectively, outpacing the best-performing alternative methods by 8.5%, 8.2%, and 12.3%. We also evaluate our approach on the MSR-VTT video-text retrieval dataset, where it delivers competitive video-to-text and text-to-video retrieval performance, while utilizing substantially less fine-tuning data compared to other methods.

Sketch Video Synthesis

AbstractUnderstanding semantic intricacies and high‐level concepts is essential in image sketch generation, and this challenge becomes even more formidable when applied to the domain of videos. To address this, we propose a novel optimization‐based framework for sketching videos represented by the frame‐wise Bézier Curves. In detail, we first propose a cross‐frame stroke initialization approach to warm up the location and the width of each curve. Then, we optimize the locations of these curves by utilizing a semantic loss based on CLIP features and a newly designed consistency loss using the self‐decomposed 2D atlas network. Built upon these design elements, the resulting sketch video showcases notable visual abstraction and temporal coherence. Furthermore, by transforming a video into vector lines through the sketching process, our method unlocks applications in sketch‐based video editing and video doodling, enabled through video composition.

Collaborative Multi-task Learning for Multi-Object Tracking and Segmentation

The advancement of computer vision has pushed visual analysis tasks from still images to the video domain. In recent years, video instance segmentation, which aims at tracking and segment multiple objects in video frames, has drawn much attention for its potential applications in various emerging areas such as autonomous driving, intelligent transportation, and smart retail. In this article, we propose an effective framework for instance-level visual analysis on video frames, which can simultaneously conduct object detection, instance segmentation, and multi-object tracking. The core idea of our method is collaborative multi-task learning which is achieved by a novel structure, named associative connections among detection, segmentation, and tracking task heads in an end-to-end learnable CNN. These additional connections allow information propagation across multiple related tasks, so as to benefit these tasks simultaneously. We evaluate the proposed method extensively on KITTI MOTS and MOTS Challenge datasets and obtain quite encouraging results.

Motion Vector-Based Self-Attention for Real-Time Human Activity Recognition in Compressed Videos: The MVViT Approach

Herein, a novel methodology is proposed for real-time recognition of human activity in a compressed domain of videos based on motion vectors and self-attention mechanism using vision transformers, and it is termed as motion vectors and vision transformers (MVViT). The videos in MPEG-4 and H.264 compression formats are considered for this study. Any video source without any prior setup could be considered by adopting the proposed method to the corresponding video codecs and camera settings. Existing algorithms for recognition of human action in a compressed video have some limitations in this regard, such as (i) requirement of keyframes at a fixed interval, (ii) usage of P frames only, and (iii) normally support single codec only. These limitations are overcome in the proposed method by using arbitrary keyframe intervals, using both P and B frames, and supporting MPEG-4 as well as H.264 codecs. The experimentation is carried out using the benchmark datasets, namely, UCF101, HMDB51, and THUMOS14, and the recognition accuracy in a compressed domain is found to be comparable to that observed in raw video data but at reduced cost of computation. The proposed MVViT method has outperformed other recent methods in terms of a lesser (61.0%) number of parameters and (63.7%) Giga Floating Point Operations Per Second (GFLOPS), while significantly improving accuracy by 0.8%, 5.9% and 16.6% for UCF101, HMDB51 and THUMOS14, respectively. Also, it is observed that the speed is increased by 8% in case of UCF101 when compared to the highest speed reported in the literature on the same dataset. The ablation study of the proposed method has been done using MVViT variants for different codecs and the performance analysis is done in comparison with the state-of-the-art network models.

A deep audio-visual model for efficient dynamic video summarization

A deep audio-visual model for efficient dynamic video summarization

ConditionVideo: Training-Free Condition-Guided Video Generation

Recent works have successfully extended large-scale text-to-image models to the video domain, producing promising results but at a high computational cost and requiring a large amount of video data. In this work, we introduce ConditionVideo, a training-free approach to text-to-video generation based on the provided condition, video, and input text, by leveraging the power of off-the-shelf text-to-image generation methods (e.g., Stable Diffusion). ConditionVideo generates realistic dynamic videos from random noise or given scene videos. Our method explicitly disentangles the motion representation into condition-guided and scenery motion components. To this end, the ConditionVideo model is designed with a UNet branch and a control branch. To improve temporal coherence, we introduce sparse bi-directional spatial-temporal attention (sBiST-Attn). The 3D control network extends the conventional 2D controlnet model, aiming to strengthen conditional generation accuracy by additionally leveraging the bi-directional frames in the temporal domain. Our method exhibits superior performance in terms of frame consistency, clip score, and conditional accuracy, outperforming other compared methods.

VadCLIP: Adapting Vision-Language Models for Weakly Supervised Video Anomaly Detection

The recent contrastive language-image pre-training (CLIP) model has shown great success in a wide range of image-level tasks, revealing remarkable ability for learning powerful visual representations with rich semantics. An open and worthwhile problem is efficiently adapting such a strong model to the video domain and designing a robust video anomaly detector. In this work, we propose VadCLIP, a new paradigm for weakly supervised video anomaly detection (WSVAD) by leveraging the frozen CLIP model directly without any pre-training and fine-tuning process. Unlike current works that directly feed extracted features into the weakly supervised classifier for frame-level binary classification, VadCLIP makes full use of fine-grained associations between vision and language on the strength of CLIP and involves dual branch. One branch simply utilizes visual features for coarse-grained binary classification, while the other fully leverages the fine-grained language-image alignment. With the benefit of dual branch, VadCLIP achieves both coarse-grained and fine-grained video anomaly detection by transferring pre-trained knowledge from CLIP to WSVAD task. We conduct extensive experiments on two commonly-used benchmarks, demonstrating that VadCLIP achieves the best performance on both coarse-grained and fine-grained WSVAD, surpassing the state-of-the-art methods by a large margin. Specifically, VadCLIP achieves 84.51% AP and 88.02% AUC on XD-Violence and UCF-Crime, respectively. Code and features are released at https://github.com/nwpu-zxr/VadCLIP.

ViLT-CLIP: Video and Language Tuning CLIP with Multimodal Prompt Learning and Scenario-Guided Optimization

Pre-trained vision-language(V-L) models such as CLIP have demonstrated impressive Zero-Shot performance in many downstream tasks. Since adopting contrastive video-text pairs methods like CLIP to video tasks is limited by its high cost and scale, recent approaches focus on efficiently transferring the image-based CLIP to the video domain. A major finding is that fine-tuning the pre-trained model to achieve strong fully supervised performance leads to low zero shot, few shot, and base to novel generalization. Instead, freezing the backbone network to maintain generalization ability weakens fully supervised performance. Otherwise, no single prompt tuning branch consistently performs optimally. In this work, we proposed a multimodal prompt learning scheme that balances supervised and generalized performance. Our prompting approach contains three sections: 1) Independent prompt on both the vision and text branches to learn the language and visual contexts. 2) Inter-modal prompt mapping to ensure mutual synergy. 3) Reducing the discrepancy between the hand-crafted prompt (a video of a person doing [CLS]) and the learnable prompt, to alleviate the forgetting about essential video scenarios. Extensive validation of fully supervised, zero-shot, few-shot, base-to-novel generalization settings for video recognition indicates that the proposed approach achieves competitive performance with less commute cost.

Deep spatiotemporal fusion network for vision-based robotic inspection of structures

Deep spatiotemporal fusion network for vision-based robotic inspection of structures

Performance Comparison of VVC, AV1, HEVC, and AVC for High Resolutions

Over the years, there has been growing interest in multimedia services, especially in the video domain, where firms and subscribers require higher resolutions, framerates, and sampling precision. This results in a huge amount of data that needs to be processed, stored, and transmitted. As a result, researchers face the challenge of developing new compression standards that can reduce the amount of data while maintaining the same quality. In this paper, the compression performance of the latest and most commonly used video codecs, namely H.266/VVC, AV1, H265/HEVC, and H.264/AVC was examined. The test set included seven sequences of various content at 8K, Ultra HD (UHD), and Full HD (FHD) resolutions, encoded to bitrates ranging from 1 to 15 Mbps for FHD and UHD resolutions and from 5 to 50 Mbps for 8K resolution. Objective quality metrics, such as peak signal-to-noise ratio (PSNR), the structural similarity index (SSIM), and video multi-method assessment fusion (VMAF) were used to measure codec performance. The results showed that H.266/VVC outperformed all other codecs, namely H.264/AVC, H.265/HEVC, and AV1, in terms of the Bjøntegaard delta (BD) model. The average bitrate savings were approximately 78% for H.266/VVC, 63% for AV1, and 53% for H.265/HEVC relative to H.264/AVC, 59% for H.266/VVC and 22% for AV1 compared to H.264/AVC, and 46% for H.266/VVC relative to AV1 (all for 8K resolution). The results also showed that codec performance varied depending on resolution, with higher resolutions showing greater efficiency for newly developed codecs, such as H.266/VVC and AV1. This confirms the fact that the H.266/VVC and AV1 codecs were primarily developed for videos at high resolutions, such as 8K and/or UHD.

Simplifying open-set video domain adaptation with contrastive learning

Simplifying open-set video domain adaptation with contrastive learning