Thanks for @sacmehta 's talent work,
I have made one change for easy using of this project,
main.py :1.add_ignoredId argument to notice which ID (in the label file) should be ignored.
2. the default value of _ignoredId is 255 (for cityscapse trainID of )
3. using torch.nn.crossEntryLoss to replace the user defined crossEntryLoss entity, and pass
_ignoredId to _ignoreIndex parameter of torch.nn.crossEntryLoss.
loaddata.py4. when check the max_value in the label file, compare it with _ignoredId first, then print some
information to notice the ignored_Id in the label file has been ignored
these changes have been verified in ubuntu 16.04 , cuda 9.0 cudnn 7.0, pytorch 1.0.0 post2,
Thanks for your time and looking for your reply.
Thanks for @sacmehta 's talent work, I have made one change for easy using of this project,
main.py : 1. add _ignoredId argument to notice which ID (in the label file) should be ignored.
2. the default value of _ignoredId is 255 (for cityscapse trainID of)
3. using torch.nn.crossEntryLoss to replace the user defined crossEntryLoss entity, and pass _ignoredId to _ignoreIndex parameter of torch.nn.crossEntryLoss.
loaddata.py 4. when check the max_value in the label file, compare it with _ignoredId first, then print some information to notice the ignored_Id in the label file has been ignored
these changes have been verified in ubuntu 16.04 , cuda 9.0 cudnn 7.0, pytorch 1.0.0 post2, Thanks for your time and looking for your reply.