pytorch-grad-cam
pytorch-grad-cam copied to clipboard
jacobgil
Are there plans to support YOLOX？
Are there plans to support YOLOX？I tried YOLOX myself and it didn't seem to work very well
Not planning, but I think it should be similar to the ideas in the YOLO/Faster RCNN tutorials.
PRs for notebooks or other examples with this are most welcome.
Thanks for your reply. I have a question, why is it better to choose model.model.model.model[-2] 23rd layer than model.model.model.model[-5] 20th layer in yolov5s? In fact, they are both pre-layer of Detection.
- result of target_layers = [model.model.model.model[23]]
- result of target_layers=[model.model.model.model[20]]
If I use YOLOX, do you have any suggestions for selecting target_layers? Thank you very much.
Hello friend, have you successfully applied thisHello friend, have you successfully applied this on yolox? If so, can you communicate with me? I am also doing relevant research.
thisHello
not yet，I have tried some target layers，but only model.backbone.dark4 looks a little bit like right. I'm not sure that's right.
OK, thank you for your response,and i find someothers,like C3_n4 in backbone, cls_conv[2], obj_conv[2] and reg_preds[2] in head may make a seemingly correct result. in addition, I found that only EigenCAM can be directly applied to YOLOX，maybe I made a mistake。
OK, thank you for your response,and i find someothers,like C3_n4 in backbone, cls_conv[2], obj_conv[2] and reg_preds[2] in head may make a seemingly correct result. in addition, I found that only EigenCAM can be directly applied to YOLOX，maybe I made a mistake。
hello friend. have u ever successfully applied this into yolox?
OK, thank you for your response,and i find someothers,like C3_n4 in backbone, cls_conv[2], obj_conv[2] and reg_preds[2] in head may make a seemingly correct result. in addition, I found that only EigenCAM can be directly applied to YOLOX，maybe I made a mistake。
hello friend. have u ever successfully applied this into yolox?how do it?
OK, thank you for your response,and i find someothers,like C3_n4 in backbone, cls_conv[2], obj_conv[2] and reg_preds[2] in head may make a seemingly correct result. in addition, I found that only EigenCAM can be directly applied to YOLOX，maybe I made a mistake。
Hello! I also found this problem. I think it should be because Yolox's head is decoupled. Can you tell me your latest progress. Thank you.
OK, thank you for your response,and i find someothers,like C3_n4 in backbone, cls_conv[2], obj_conv[2] and reg_preds[2] in head may make a seemingly correct result. in addition, I found that only EigenCAM can be directly applied to YOLOX，maybe I made a mistake。
Hello! I also found this problem. I think it should be because Yolox's head is decoupled. Can you tell me your latest progress. Thank you.
target_layers = [model.backbone.backbone.dark5] target_layers = [model.head.obj_preds[2], model.head.reg_preds[2], model.head.cls_preds[2]] I set it up like this，
OK, thank you for your response,and i find someothers,like C3_n4 in backbone, cls_conv[2], obj_conv[2] and reg_preds[2] in head may make a seemingly correct result. in addition, I found that only EigenCAM can be directly applied to YOLOX，maybe I made a mistake。
Hello! I also found this problem. I think it should be because Yolox's head is decoupled. Can you tell me your latest progress. Thank you.
target_layers = [model.backbone.backbone.dark5] target_layers = [model.head.obj_preds[2], model.head.reg_preds[2], model.head.cls_preds[2]] I set it up like this，
Thank you for your reply! The target layers I selected are the same as yours, but when I output the thermal map, I found that the target area is not highlighted but the background area is highlighted. Do you know why? Look forward to your reply!!!
OK, thank you for your response,and i find someothers,like C3_n4 in backbone, cls_conv[2], obj_conv[2] and reg_preds[2] in head may make a seemingly correct result. in addition, I found that only EigenCAM can be directly applied to YOLOX，maybe I made a mistake。
Hello! I also found this problem. I think it should be because Yolox's head is decoupled. Can you tell me your latest progress. Thank you.
target_layers = [model.backbone.backbone.dark5] target_layers = [model.head.obj_preds[2], model.head.reg_preds[2], model.head.cls_preds[2]] I set it up like this，
Thank you for your reply! The target layers I selected are the same as yours, but when I output the thermal map, I found that the target area is not highlighted but the background area is highlighted. Do you know why? Look forward to your reply!!!
Maybe you can try a different setup， mine turned out like this
thisHello
not yet，I have tried some target layers，but only model.backbone.dark4 looks a little bit like right. I'm not sure that's right.
Hello, I found the same situation in my experiment and couldn't get satisfactory results after changing multiple settings. Could you please tell me how you solved it?
