Closed Godatplay closed 5 years ago
@Godatplay Hi, the project is intended to test on HM3.6M of original version and eccv18 challenge. We do not recommend to test on MPII, because we do not include related codes. If you insist, you could write your own _mpiiDataset function refering _hm36Dataset, and pass all necessary parameters when call it.
Good to know, thank you!
Stop wasting time. Integral loss does not work on MPII dataset at all. The actual effect is totally not as good as that described in the paper.
I'm trying to test on MPII with a custom config set to my data path and using the test.py found in integral_human_pose. I'm getting this error:
'mpii_Dataset' is not defined
because of course you're using mpii_hm36_Dataset. Is this just a bug/oversight? Or am I missing something? Changing my dataset name to that in the config does not work, and changing the import in test.py to import as mpii_Dataset also does not work, producing this error:TypeError: __init__() missing 1 required positional argument: 'label_config'
How should I test on MPII?