MACHINE PERCEPTION AND ROBOTICS GROUP

機械知覚&ロボティクスグループ
中部大学

Attention branch network (ABN) achieves high accuracy by visualizing the attention area of the network during inference and utilizing it in the recognition process. Meanwhile, if the attention area does not highlight the target object to be recognized, it may causes failure recognition. To solve this problem, a fine-tuning method of the ABN using modified attention maps by human knowledge has been proposed. However, this method takes human and time costs because the attention map must be modified by humans. In this paper, we propose a method that automatically optimize the attention map by introducing an attention mining branch to the ABN.