Closed ishmamahmed closed 8 months ago
Hi, thanks for your attention to our work. The value '0' represents the background class, which is not calculated for the mIoU and loss. To be specific, this line process the input labels. In NYU dataset, the background class tends to be the borders between the objects and the edge of the images, as shown in below (black is the background class):
Thank you very much for your clarification!!
After subtracting 1 from the ground truth, the background (which was previously 0) becomes -1.
And I see you have ignored -1 in the cross-entropy loss using ignore_index:
The data format of the label is uint8 (0~255), so '0 - 1' is 255.
Hello, thank you for sharing the code of this great work!!
I have a query about the NYU dataset.
The NYU dataset has 40 classes. So, in the final output layer of all of your decoders, num_classes is set to 40.
When I check the pixel values of the ground truths (0.png, 1.png, etc) inside the NYUDepthv2/Label/ directory, the pixel values range from 0 to 40. This indicates there are 41 classes.
So can you kindly tell me how have you dealt with the extra class in the ground truth labels? In my code, this discrepency causes error in the cross-entropy loss function.
Thank you.