Text-to-video (T2V) generation has made tremendous progress in generating complicated scenes based on texts. However, human-object interaction (HOI) often cannot be precisely generated by current T2V models due to the lack of large-scale videos with accurate captions for HOI. To address this issue, we introduce HOIGen-1M, the first large-scale dataset for HOI Generation, consisting of over one million high-quality videos collected from diverse sources. In particular, to guarantee the high quality of videos, we first design an efficient framework to automatically curate HOI videos using the powerful multimodal large language models (MLLMs), and then the videos are further cleaned by human annotators. Moreover, to obtain accurate textual captions for HOI videos, we design a novel video description method based on a Mixture-of-Multimodal-Experts (MoME) strategy that not only generates expressive captions but also eliminates the hallucination by individual MLLM. Further-more, due to the lack of an evaluation framework for gen-erated HOI videos, we propose two new metrics to assess the quality of generated videos in a coarse-to-fine manner. Extensive experiments reveal that current T2V models struggle to generate high-quality HOI videos and confirm that our HOIGen-1M dataset is instrumental for improving HOI video generation.

Spatio-Temporal Video Grounding (STVG) aims at localizing the spatio-temporal tube of a specific object in an untrimmed video given a free-form natural language query. As the annotation of tubes is labor intensive, researchers are motivated to explore weakly supervised approaches in recent works, which usually results in significant performance degradation. To achieve a less expensive STVG method with acceptable accuracy, this work investigates the "single-frame supervision" paradigm that requires a single frame labeled with a bounding box within the temporal boundary of the fully supervised counterpart as the supervisory signal. Based on the characteristics of the STVG problem, we propose a Two-Stage Multiple Instance Learning (T-SMILE) method, which creates pseudo labels by expanding the annotated frame to its contextual frames, thereby establishing a fully-supervised problem to facilitate further model training. The innovations of the proposed method are three-folded, including 1) utilizing multiple instance learning to dynamically select instances in positive bags for the recognition of starting and ending timestamps, 2) learning highly discriminative query features by incorporating spatial prior constraints in cross-attention, and 3) designing a curriculum learning-based strategy that iterative assigns dynamic weights to spatial and temporal branches, thereby gradually adapting to the learning branch with larger difficulty. To facilitate future research on this task, we also contribute a large-scale benchmark containing 12,469 videos on complex scenes with single-frame annotation. The extensive experiments on two benchmarks demonstrate that T-SMILE significantly outperforms all weakly-supervised methods. Remarkably, it also performs better than some fully-supervised methods associated with much more annotation labor costs.

Multi-modal human action segmentation is a critical and challenging task with a wide range of applications. Nowadays, the majority of approaches concentrate on the fusion of dense signals (i.e., RGB, optical flow, and depth maps). However, the potential contributions of sparse IoT sensor signals, which can be crucial for achieving accurate recognition, have not been fully explored. To make up for this, we introduce a S parse s i gnal- g uided Transformer ( SigFormer ) to combine both dense and sparse signals. We employ mask attention to fuse localized features by constraining cross-attention within the regions where sparse signals are valid. However, since sparse signals are discrete, they lack sufficient information about the temporal action boundaries. Therefore, in SigFormer, we propose to emphasize the boundary information at two stages to alleviate this problem. In the first feature extraction stage, we introduce an intermediate bottleneck module to jointly learn both category and boundary features of each dense modality through the inner loss functions. After the fusion of dense modalities and sparse signals, we then devise a two-branch architecture that explicitly models the interrelationship between action category and temporal boundary. Experimental results demonstrate that SigFormer outperforms the state-of-the-art approaches on a multi-modal action segmentation dataset from real industrial environments, reaching an outstanding F1 score of 0.958. The codes and pre-trained models have been made available at https://github.com/LIUQI-creat/SigFormer .

Action recognition from videos, i.e., classifying a video into one of the pre-defined action types, has been a popular topic in the communities of artificial intelligence, multimedia, and signal processing. However, existing methods usually consider an input video as a whole and learn models, e.g., Convolutional Neural Networks (CNNs), with coarse video-level class labels. These methods can only output an action class for the video, but cannot provide fine-grained and explainable cues to answer why the video shows a specific action. Therefore, researchers start to focus on a new task, Part-level Action Parsing (PAP), which aims to not only predict the video-level action but also recognize the frame-level fine-grained actions or interactions of body parts for each person in the video. To this end, we propose a coarse-to-fine framework for this challenging task. In particular, our framework first predicts the video-level class of the input video, then localizes the body parts and predicts the part-level action. Moreover, to balance the accuracy and computation in part-level action parsing, we propose to recognize the part-level actions by segment-level features. Furthermore, to overcome the ambiguity of body parts, we propose a pose-guided positional embedding method to accurately localize body parts. Through comprehensive experiments on a large-scale dataset, i.e., Kinetics-TPS, our framework achieves state-of-the-art performance and outperforms existing methods over 31.10% ROC score.

Deep neural networks have achieved remarkable success for video-based action recognition. However, most of existing approaches cannot be deployed in practice due to the high computational cost. To address this challenge, we propose a new real-time convolutional architecture, called Temporal Convolutional 3D Network (T-C3D), for action representation. T-C3D learns video action representations in a hierarchical multi-granularity manner while obtaining a high process speed. Specifically, we propose a residual 3D Convolutional Neural Network (CNN) to capture complementary information on the appearance of a single frame and the motion between consecutive frames. Based on this CNN, we develop a new temporal encoding method to explore the temporal dynamics of the whole video. Furthermore, we integrate deep compression techniques with T-C3D to further accelerate the deployment of models via reducing the size of the model. By these means, heavy calculations can be avoided when doing the inference, which enables the method to deal with videos beyond real-time speed while keeping promising performance. We validate our approach by studying its action representation performance on four benchmarks over three different tasks. Our method achieves clear improvements on UCF101 action recognition benchmark against the state-of-the-art real-time methods by 5.4% in terms of accuracy and 2 times faster in terms of inference speed with a less than 5MB storage model. The source code and the pre-trained models are publicly available at https://github.com/tc3d.

Existing sketch-analysis work studies sketches depicting static objects or scenes. In this work, we propose a novel cross-modal retrieval problem of fine-grained instance-level sketch-based video retrieval (FG-SBVR), where a sketch sequence is used as a query to retrieve a specific target video instance. Compared with sketch-based still image retrieval, and coarse-grained category-level video retrieval, this is more challenging as both visual appearance and motion need to be simultaneously matched at a fine-grained level. We contribute the first FG-SBVR dataset with rich annotations. We then introduce a novel multi-stream multi-modality deep network to perform FG-SBVR under both strong and weakly supervised settings. The key component of the network is a relation module, designed to prevent model overfitting given scarce training data. We show that this model significantly outperforms a number of existing state-of-the-art models designed for video analysis.

Anomaly detection in surveillance videos, as a special case of video-based action recognition, has been of increasing interest in multimedia community and public security. Action recognition in videos faces some challenges, such as cluttered background, illumination conditions. Besides these above difficulties, detecting anomaly in surveillance videos has several unique problems to be solved. For example, the lack of sufficient training samples is one of the main challenges for detecting anomalies in surveillance videos. In this paper, we propose to utilize transfer learning to leverage the good results from action recognition for anomaly detection in surveillance videos. More specially, we explore some techniques based on action recognition models from the following aspects: training samples, temporal modules for action recognition, network backbones. We draw some conclusions. First, more training samples from surveillance videos lead to higher classification accuracy. Second, stronger temporal modules designed for recognizing action and deeper networks do not achieve better results. This conclusion is reasonable since deeper networks tend to over-fitting, especially for the small-scale training set. Besides, to distinguish the hard examples from normal activities, we separately train a neural network to classify the hard category and normal events. Then we fuse the binary network and previous network to generate the final prediction for general anomaly detection. On the benchmarks of CitySCENE, our framework achieves promising performance and obtains the first prize for general anomaly detection and the second prize for specific anomaly detection.

Anomaly detection in surveillance videos, as a special case of video-based action recognition, is an important topic in multimedia community and public security. Currently, most of the state-of-the-art methods utilize deep learning to recognize the patterns of anomaly or action. However, whether deep neural networks really learn the essence of the anomaly or just remember the background is an important but often neglected problem. In this paper, we develop a series of experiments to validate the existence of background-bias phenomenon, which makes deep networks tend to learn the background information rather than the pattern of anomalies to recognize abnormal behavior. To solve it, we first re-annotate the largest anomaly detection dataset and design a new evaluation metric to measure whether the models really learn the essence of anomalies. Then, we propose an end-to-end trainable, anomaly-area guided framework, where we design a novel region loss to explicitly drive the network to learn where is anomalous region. Besides, given very deep networks and scarce training data for anomaly, our architecture is trained with a meta learning module to prevent severe overfitting. Extensive experiments on the benchmark show that our approach outperforms other methods on both the previous and our proposed evaluation metrics through reducing the influence of the background information.

Video-based action recognition with deep neural networks has shown remarkable progress. However, most of the existing approaches are too computationally expensive due to the complex network architecture. To address these problems, we propose a new real-time action recognition architecture, called Temporal Convolutional 3D Network (T-C3D), which learns video action representations in a hierarchical multi-granularity manner. Specifically, we combine a residual 3D convolutional neural network which captures complementary information on the appearance of a single frame and the motion between consecutive frames with a new temporal encoding method to explore the temporal dynamics of the whole video. Thus heavy calculations are avoided when doing the inference, which enables the method to be capable of real-time processing. On two challenging benchmark datasets, UCF101 and HMDB51, our method is significantly better than state-of-the-art real-time methods by over 5.4% in terms of accuracy and 2 times faster in terms of inference speed (969 frames per second), demonstrating comparable recognition performance to the state-of-the-art methods. The source code for the complete system as well as the pre-trained models are publicly available at https://github.com/tc3d.