With the growing popularity of high-resolution (HR) video and the continuous growth of network bandwidth, the challenge of object removal detection in HR videos has attracted significant attention. Expert forgers leverage the rich detail in HR videos for meticulous pixel manipulation and apply sophisticated postprocessing techniques to hide high-frequency artifacts, thereby making forgery detection and localization more difficult when existing schemes are used. Additionally, the end-to-end framework simplifies the detection and localization process, which has not been considered in previous work. To solve the above issues, a spatiotemporal encoder−decoder network (SEDN) is proposed for end-to-end object removal forgery detection in HR videos. In the SEDN, a new model composed of a 3D asymmetric dual-stream network (3D-ADSN) and Transformer is proposed. The 3D-ADSN is utilized as the encoder, which fully integrates the high-frequency and low-frequency spatiotemporal information of videos. Transformer is utilized as the decoder to capture the global structure spatiotemporal information of the long-range feature sequence obtained by the encoder. This network combination successfully achieves simultaneous detection in the temporal and spatial domains without any additional postprocessing calculations. The experimental results demonstrate the better performance of the SEDN at different resolutions.

Traditional video action detectors typically adopt the two-stage pipeline, where a person detector is first employed to generate actor boxes and then 3D RoIAlign is used to extract actor-specific features for action recognition. This detection paradigm requires multi-stage training and inference, and the feature sampling is only constrained inside the box, failing to effectively leverage richer context information outside. Recently, several query-based action detectors have been proposed to predict action instances in an end-to-end manner. However, they still lack adaptability in feature sampling and decoding, thus suffering from the issues of inferior performance or slower convergence. In this paper, we propose two core designs for a more flexible one-stage sparse action detector. First, we present a query-based adaptive feature sampling module, which endows the detector with the flexibility of mining a group of discriminative features from the entire spatio-temporal domain. Second, we devise a decoupled feature mixing module, which dynamically attends to and mixes video features along the spatial and temporal dimensions respectively for better feature decoding. Based on these designs, we instantiate two detection pipelines, that is, STMixer-K for keyframe action detection and STMixer-T for action tubelet detection. Without bells and whistles, our STMixer detectors obtain the state-of-the-art results on five challenging spatio-temporal action detection benchmarks for keyframe action detection or action tube detection.

Traditional video action detectors typically adopt the two-stage pipeline, where a person detector is first employed to generate actor boxes and then 3D RoIAlign is used to extract actor-specific features for classification. This detection paradigm requires multi-stage training and inference, and cannot capture context information outside the bounding box. Recently, a few query-based action detectors are proposed to predict action instances in an end-to-end manner. However, they still lack adaptability in feature sampling and decoding, thus suffering from the issues of inferior performance or slower convergence. In this paper, we propose a new one-stage sparse action detector, termed STMixer. STMixer is based on two core designs. First, we present a query-based adaptive feature sampling module, which endows our STMixer with the flexibility of mining a set of discriminative features from the entire spatiotemporal domain. Second, we devise a dual-branch feature mixing module, which allows our STMixer to dynamically attend to and mix video features along the spatial and the temporal dimension respectively for better feature decoding. Coupling these two designs with a video backbone yields an efficient end-to-end action detector. Without bells and whistles, our STMixer obtains the state-of-the-art results on the datasets of AVA, UCF101-24, and JHMDB.

Action spotting and classification consist in detecting the exact moments at which events occur in long videos. The current mainstream spotting practices generally use a two-stage pipeline that performs feature collection and integration, then salient action detection and postprocessing. Following that, we present SpotFormer, a simple yet effective framework, capable of precise action spotting. Specifically, we employ several most advanced backbone networks as auxiliary feature extractors, and reduce feature dimensionality in a straightforward and efficient way. The frame-wise features are fed into a transformer-based spotting network devised to leverage spatiotemporal information. We obtain 0.609 tight mAP score via model ensemble and achieve the state-of-the-art performance on the SoccerNet-v2 dataset.

Abundant tourism user-generated content (UGC) contains a wealth of cognitive and emotional information, providing valuable data for building destination images that depict tourists’ experiences and appraisal of the destinations during the tours. In particular, multiple destination images can assist tourism managers in exploring the commonalities and differences to investigate the elements of interest of tourists and improve the competitiveness of the destinations. However, existing methods usually focus on the image of a single destination, and they are not adequate to analyze and visualize UGC to extract valuable information and knowledge. Therefore, we discuss requirements with tourism experts and present MDIVis, a multi-level interactive visual analytics system that allows analysts to comprehend and analyze the cognitive themes and emotional experiences of multiple destination images for comparison. Specifically, we design a novel sentiment matrix view to summarize multiple destination images and improve two classic views to analyze the time-series pattern and compare the detailed information of images. Finally, we demonstrate the utility of MDIVis through three case studies with domain experts on real-world data, and the usability and effectiveness are confirmed through expert interviews.