Survey of vision transformers for action recognition
Survey of vision transformers for action recognition
Vision Transformers for Action Recognition: A Survey
arXiv paper abstract https://arxiv.org/abs/2209.05700v1
arXiv PDF paper https://arxiv.org/pdf/2209.05700v1.pdf
… provides the first comprehensive survey of vision transformer techniques for action recognition.
… analyze and summarize the existing and emerging literature in this direction while highlighting the popular trends in adapting transformers for action recognition.
… literature review provides suitable taxonomies for action transformers based on their architecture, modality, and intended objective.
… explore the techniques to encode spatio-temporal data, dimensionality reduction, frame patch and spatio-temporal cube construction, and various representation methods.
… investigate the optimization of spatio-temporal attention in transformer layers to handle longer sequences, typically by reducing the number of tokens in a single attention operation.
… investigate different network learning strategies, such as self-supervised and zero-shot learning, along with their associated losses for transformer-based action recognition …
Stay up to date. Subscribe to my posts https://morrislee1234.wixsite.com/website/contact
Web site with my other posts by category https://morrislee1234.wixsite.com/website
LinkedIn https://www.linkedin.com/in/morris-lee-47877b7b
