I2R's team won 1st in the 4th UG2+ Prize Challenge Track 2.1 (CVPR 2021)
21 Jun 2021
The A*StarTrek team comprising of Jin Ruibing, Chen Zhenghua, Wu Keyu, Hou Yubo, Wu Min, from I2R’s Machine Intellection Department, and Yang Zaifeng from IHPC competed in the Track 2.1: Fully Supervised Action Recognition in the Dark Challenge, held in tandem with CVPR 2021.
It was a tough fight against 35 Teams from all around the world, which include companies like Samsung AI, Tencent, DeepBlue Technology.
The teams have to train the system to recognise actions within the given dark environment videos. Such action recognition are difficult as extensive information included in videos is damaged under dark condition. Traditional training strategies are rendered useless in such situation.
Despite that, using multi-model based zigzag learning method for action recognition, the team was able to adapt the multi-model based pseudo ground truth (PGT) generation and achieved an accuracy of 93.7161% in this challenge and took home 1st Place.