search

JimmySuen / integral-human-pose

Integral Human Pose Regression
MIT License
471 stars 76 forks source link

Error testing on COCO #14

Closed Godatplay closed 5 years ago

Godatplay commented 5 years ago

I'm getting errors while trying to test on the COCO dataset. In order to attempt this, so far I've created a custom config with my paths to the data, added importing coco to all_dataset.py, added coco_Dataset to loader.py (using same template as hm36_Dataset), and am importing coco_Dataset in the test.py found in integral_human_pose. When I try to run that test script, I get this error: size mismatch for module.head.features.9.bias: copying a param of torch.Size([1224]) from checkpoint, where the shape is torch.Size([1296]) in current model. (and one for weights with the same length mismatch) Do I have the wrong version of the COCO dataset? I currently have val2017 and annotations.

lck1201 commented 5 years ago

@Godatplay Hi, the problem shall in your config file. You build a different network using different network configs. Please check the config.train.patch_height, config.train.patch_width, and network. Refer to here

lck1201 commented 5 years ago

Close issue, re-open if you have further questions