High-resolution rectified gradient-based visual explanations for weakly supervised segmentation
作者：
Highlights：
• A high-resolution rectified gradient-based CAM with bounding box is proposed.
• HRCAM-BB separates gradient from the feature map for improvement of visualization.
• LeakyReLU is applied to rectify the gradient of HRCAM-BB for improvement.
• Selected ReLU layers are removed in backpropagation for improvement.
• HRCAM-BB achieves 61.9 mIOU for AP50 on Pascal test set for WSIS.
摘要
•A high-resolution rectified gradient-based CAM with bounding box is proposed.•HRCAM-BB separates gradient from the feature map for improvement of visualization.•LeakyReLU is applied to rectify the gradient of HRCAM-BB for improvement.•Selected ReLU layers are removed in backpropagation for improvement.•HRCAM-BB achieves 61.9 mIOU for AP50 on Pascal test set for WSIS.
论文关键词：
论文评审过程：Received 25 October 2021, Revised 9 March 2022, Accepted 19 April 2022, Available online 20 April 2022, Version of Record 3 May 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.108724
