Review for NeurIPS paper: Comprehensive Attention Self-Distillation for Weakly-Supervised Object Detection

I am worried that this may also be the case for WSOD. I observe many bad signals in this paper. For example, there are many, many hyperparameters in this method.