ISCSS-FangTeam / ACGC

code for paper:Improved Pseudomasks Generation for Weakly Supervised Building Extraction From High-Resolution Remote Sensing Imagery
MIT License
2 stars 0 forks source link

about the conf_fg_thres and conf_bg_thres #1

Open Paddle15 opened 4 months ago

Paddle15 commented 4 months ago

Hello, I would like to ask a question.I would like to know how you calculated the conf_fg_thres and conf_bg_thres before the code in cam_to_ir_labels.py, correct me if I am wrong. After running eval_cam.py, you get some thresholds which I think is conf_bg_thres. so how did you get the conf_fg_thres value? It would be helpful if you could explain how these thresholds are calculated.

labiao commented 1 month ago

Sorry for the late reply. The conf_fg_thres and conf_bg_thres are used to select the foreground and background regions in the CAM. After running eval_cam.py, we select the threshold that gives the best mIoU. Based on this threshold, I then empirically set conf_fg_thres to the threshold + 0.05 and conf_bg_thres to the threshold - 0.05. Additionally, I wanted to share some exciting news: I recently had a new paper accepted in the TGRS journal, where we have further improved the performance of building extraction. The code will be public soon. Please stay tuned for more details!