C-WSL: Count-guided Weakly Supervised Localization

We introduce a count-guided weakly supervised localization (C-WSL) framework with per-class object count as an additional form of image-level supervision to improve weakly supervised localization (WSL). C-WSL uses a simple count-based region selection algorithm to select highquality regions, each of which covers a single object instance at training time, and improves WSL by training with the selected regions. To demonstrate the effectiveness of CWSL, we integrate object count supervision into two WSL architectures and conduct extensive experiments on Pascal VOC2007 and VOC2012. Experimental results show that C-WSL leads to large improvements in WSL detection performance and that the proposed approach significantly outperforms the state-of-the-art methods.