OmViD: Omni-supervised active learning for video action detection

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Rana, Aayush, Kumar, Akash, Vineet, Vibhav, Rawat, Yogesh S
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866916908813516800
author Rana, Aayush
Kumar, Akash
Vineet, Vibhav
Rawat, Yogesh S
author_facet Rana, Aayush
Kumar, Akash
Vineet, Vibhav
Rawat, Yogesh S
contents Video action detection requires dense spatio-temporal annotations, which are both challenging and expensive to obtain. However, real-world videos often vary in difficulty and may not require the same level of annotation. This paper analyzes the appropriate annotation types for each sample and their impact on spatio-temporal video action detection. It focuses on two key aspects: 1) how to obtain varying levels of annotation for videos, and 2) how to learn action detection from different annotation types. The study explores video-level tags, points, scribbles, bounding boxes, and pixel-level masks. First, a simple active learning strategy is proposed to estimate the necessary annotation type for each video. Then, a novel spatio-temporal 3D-superpixel approach is introduced to generate pseudo-labels from these annotations, enabling effective training. The approach is validated on UCF101-24 and JHMDB-21 datasets, significantly cutting annotation costs with minimal performance loss.
format Preprint
id arxiv_https___arxiv_org_abs_2508_13983
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle OmViD: Omni-supervised active learning for video action detection
Rana, Aayush
Kumar, Akash
Vineet, Vibhav
Rawat, Yogesh S
Computer Vision and Pattern Recognition
Video action detection requires dense spatio-temporal annotations, which are both challenging and expensive to obtain. However, real-world videos often vary in difficulty and may not require the same level of annotation. This paper analyzes the appropriate annotation types for each sample and their impact on spatio-temporal video action detection. It focuses on two key aspects: 1) how to obtain varying levels of annotation for videos, and 2) how to learn action detection from different annotation types. The study explores video-level tags, points, scribbles, bounding boxes, and pixel-level masks. First, a simple active learning strategy is proposed to estimate the necessary annotation type for each video. Then, a novel spatio-temporal 3D-superpixel approach is introduced to generate pseudo-labels from these annotations, enabling effective training. The approach is validated on UCF101-24 and JHMDB-21 datasets, significantly cutting annotation costs with minimal performance loss.
title OmViD: Omni-supervised active learning for video action detection
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.13983