Scene Recognition in Short Video with Multi-Resolution CNNs

In order to solve the problems of scene recognition in short videos, this paper proposes a deep fusion network based on VGGNet. Firstly, VGGNet16 is used to learn global features, and VGGNet19 is used to learn images in details. After that the learning features are fused by means of weighted averaging; In the public dataset 2017-AI-Challenger-scene-classification, the result of top3 is 92.2% and the top3 of the Charades short video dataset has achieved 78.9%, which proves that the proposed method has a good performance in scene recognition.