Do less and achieve more: Training CNNs for action recognition utilizing action images from the Web
作者：
Highlights：
• We collect three large web action image datasets.
• We verify that web action images are complementary to training videos by extensive experiments.
• We show both filtered and unfiltered web action images are complementary to training videos.
• We show usefulness of web action images in solving an artifact of finetuning CNN model.
摘要
•We collect three large web action image datasets.•We verify that web action images are complementary to training videos by extensive experiments.•We show both filtered and unfiltered web action images are complementary to training videos.•We show usefulness of web action images in solving an artifact of finetuning CNN model.
论文关键词：Action recognition,Convolutional neural networks,Deep learning,Videos,Images
论文评审过程：Received 15 March 2016, Revised 13 January 2017, Accepted 20 January 2017, Available online 1 February 2017, Version of Record 22 April 2017.
论文官网地址：https://doi.org/10.1016/j.patcog.2017.01.027
