search

wmcnally / kapao

KAPAO is an efficient single-stage human pose estimation model that detects keypoints and poses as objects and fuses the detections to predict human poses.
GNU General Public License v3.0
753 stars 103 forks source link

Albumentations leads to batch concatenating fail #64

Closed V63A closed 2 years ago

V63A commented 2 years ago

Hi, thank you for the great work!

I noticed that you comment out the package 'albumentations' in requirement.txt, but I guess you just follow the setting of yolov5. Have you ever trained with this package? I tried to train with it, and got this error:

image

It failed to concatenate labels due to the mismatched dimension, and I found out that the shape of variable 'labels' of an image without any label information will be changed after albumentations calcuation(line 594 in utils/datasets.py ), from (0,56) to (0,).

So, if there is any recommended modified, or manually coverts the shape back with value 0 is ok?

wmcnally commented 2 years ago

No I have not tried training with this package. You may have to modify the Albumentations class in utils.augmentations to get it working.