Uncertainty Estimation via Response Scaling for Pseudo-Mask Noise Mitigation in Weakly-Supervised Semantic Segmentation

Weakly-Supervised Semantic Segmentation (WSSS) segments objects without heavy burden of dense annotation. While as a price, generated pseudo-masks exist obvious noisy pixels, which result in sub-optimal segmentation models trained over these pseudo-masks. But rare studies notice or work on this problem, even pixels are inevitable after their improvements pseudo-mask. So we try to improve WSSS the aspect noise mitigation. And observe that many high confidences, especially when response range is too wide narrow, presenting an uncertain status. Thus, paper, simulate variations by scaling prediction map multiple times for uncertainty estimation. The then used weight loss mitigate supervision signals. We call method URN, abbreviated from Uncertainty estimation via Response Noise Experiments validate benefits and our achieves state-of-the-art results at 71.2% 41.5% PASCAL VOC 2012 MS COCO 2014 respectively, extra like saliency detection. Code available https://github.com/XMed-Lab/URN.