I have the same question. Could you tell me the idea you draw it?

@DingShengLin May be you can use CAM to output the spatial sensitive heatmap.

Hi, we obtain the sensitive heatmap depends on the accuracy of the classification and regression of the region proposal in this location. Specifically, given a feature map H in the detector, we can obtain the corresponding region proposals in each coordinate in H and then forward them into the detection head to obtain the classification score and IoU of regression. Take the classification sensitive heatmap for example, we assign the sensitive score S*C for each coordinate in a proposal via its classification score C. The central point will be assigned a value of S=1, where S is computed by a 2-D Gaussian function, fading from the center of the suggestion box to the edge. By computing all of the sensitive scores for each coordinate, we can draw this spatial sensitive heatmap. The regression sensitive heatmap can be obtained with the same method.

@songguanglu Thank you for your explanation. Is it convenient to make the code public?