Paper Info
Title
Multi-teacher knowledge distillation for compressed video action recognition based on deep learning.
Abstract
Recently, Convolutional Networks have great progress in classifying images. While action recognition is different from still image classification, video data contains temporal information which plays an important role in video understanding. Currently most CNN-based approaches for action recognition has excessive computational costs, an explosion of parameters and computation time. The most efficient method currently trained a deep network directly on the compressed video contains the motion information. However, this method has a large number of parameters. We propose a multi-teacher knowledge distillation framework for compressed video action recognition to compress this model. With this framework, the model is compressed by transferring the knowledge from multiple teachers to a single small student model. With multi-teacher knowledge distillation, students learn better than single-teacher knowledge distillation. Experiments show that we can reach a 2.4 ×  compression rate in number of parameters and 1.2 ×  computation reduction with 1.79% loss of accuracy on the UCF-101 dataset and 0.35% loss of accuracy on the HMDB51 dataset.
Year
DOI
Venue
2020
10.1016/j.sysarc.2019.101695
Journal of Systems Architecture
Keywords
Field
DocType
Deep convolutional model compression,Action recognition,Knowledge distillation,Transfer learning
Data compression ratio,Computer science,Action recognition,Real-time computing,Distillation,Artificial intelligence,Deep learning,Contextual image classification,Machine learning,Computation
Journal
Volume
ISSN
Citations 
103
1383-7621
2
PageRank 
References 
Authors
0.36
0
2
Name
Order
Citations
PageRank
Meng-Chieh Wu140.72
Ching-Te Chiu230438.60