Closed 872544050 closed 4 years ago
feinanshan
commented
4 years ago Hi, can you check whether the trained model in loaded properly by changing the ''False'' into ''True'' in https://github.com/feinanshan/TDNet/blob/master/Training/validate.py#L53
872544050
commented
4 years ago I have read your code carefully, and I understand almost all the parameters that can be changed. If I change this parameter to true, the following error will occur:
Traceback (most recent call last):
File "/mnt/guhuayu/TDNetxin/Training/validate.py", line 153, in
Please correct my mistake, thank you. (Among them, 51 lines of 'psp_path 'parameter seems to be redundant, I deleted it)
feinanshan
commented
4 years ago Hi please try with replacing ''state = torch.load(cfg["validating"]["resume"])#["model_state"]'' into ''state = torch.load(cfg["validating"]["resume"])["model_state"]'' in line https://github.com/feinanshan/TDNet/blob/master/Training/validate.py#L52
872544050
commented
4 years ago Dear author, thank you very much for your advice. There is no problem in the verification phase. However, the same weight is in the test.py There are still problems in the document:
Loading pretrained model from './checkpoint/td4_psp_cityscapes_best_model.pkl'
Traceback (most recent call last):
File "/mnt/guhuayu/TDNetxin/Testing/test.py", line 95, in
Kindly give me your advice.
feinanshan
commented
4 years ago Hi at this line of code https://github.com/feinanshan/TDNet/blob/3f8b5378fcc7f97c26b3760ddaf3d4402cf477d1/Testing/model/pspnet/td2_psp50.py#L162 Change ''model_state = torch.load(self.psp_path)'' into ''model_state = torch.load(self.psp_path)["model_state"]''
or you can extract the item "model_state" from the trained model, and save it as a unique file for the verification phase.
872544050
commented
4 years ago Dear author, thank you very much for your advice.
AlbertHuyb
commented
2 years ago Dear author, thank you very much for your advice.
Did you reproduce the validation result on cityscapes?
I use the provided td4-psp18 model, but only get around 50% mIoU on val data.
AvrahamRaviv
commented
1 year ago Dear author, thank you very much for your advice.
Did you reproduce the validation result on cityscapes?
I use the provided td4-psp18 model, but only get around 50% mIoU on val data.
Hi, here the same... Could you reproduce? Thanks
Dear author, for the same cityscape dataset, my mIOU in the training phase can reach more than 0.8, but in the test and verification phase, the mIOU of this weight is only 0.009. Which one of my settings is wrong? I've been training many times and I've had this problem all the time.