PSPNet-tensorflow
PSPNet-tensorflow copied to clipboard
hellochick
evaluation mIoU: 0.00404 on cityscapes val dataset
Hi, @hellochick ,
I had run the evaluation command for cityscapes val dataset using your provided trained model. However, the mIoU is too low (0.00404) which seems something wrong in evaluate.py. As the inference result using the same model gives the same output as your example. Would you give help to check your evaluate.py again?
THX!
Hi @amiltonwong, I faced this issue initially. The output given by the network is the trainIds (0 - 18). You need to convert these to Label Ids (0 - 33) This mapping is given in the labels.py script provided in cityscapes scripts.
Hi @amiltonwong , have you resolved your problem? I also encountered a similar problem on ADE20K val dataset. The mIoU is also too low.
Not yet, a conversion script for label id is required to perform correct evaluation. I'll work on it then.
Hey guys, you should donwload the dataset from here instead of this one. Maybe that can solve your questions.
@hellochick . Thank you very much! I have solved my problem by downloading a new version dataset ade20k from the Internet address you provided. Thank again!
Hi @amiltonwong, I faced this issue initially. The output given by the network is the trainIds (0 - 18). You need to convert these to Label Ids (0 - 33) This mapping is given in the labels.py script provided in cityscapes scripts.
@rydeldcosta I have the same problem, and the above approaches are used for ADE20K. But, how to evaluate cityscape? I follow you advice and run the code: "./cityscapesscripts/helpers/labels.py" but only got: `Example usages: ID of label 'car': 26 Category of label with ID '26': vehicle Name of label with trainID '0': road
Process finished with exit code 0` I confused that, how to use "./cityscapesscripts/helpers/labels.py", or change "evaluate.py" in this repo somewhere ? @hellochick
THX
Did you get a solution? I met the problem too,THX @amiltonwong @hellochick @rydeldcosta
