Self-Supervised Marine Video Analysis via Siamese Network
Liang Ju, Jihan Song, Qianqian Li, Zhensheng Shi, Zhaorui Gu, Haiyong Zheng, Bing Zheng · OCEANS 2021: San Diego – Porto · 2021
At present, advanced equipment has provided strong data support for marine scientific research. However, it is unrealistic to rely on manpower analysis to analyze the huge amount of data. Therefore, it is an effective and good method to use computer vision method to automatically identify and analyze marine video. In this paper, a self-supervised learning method based on siamese network is designed to learn the effective visual representation in marine unlabeled video, and the model is transferred to three downstream tasks: marine organism action recognition, marine organism detection, and marine scene recognition. We are on the latest dataset to experiment to evaluation the effectiveness of our method. Experimental results show that our method has certain competitiveness and effectiveness in the three downstream tasks.