Encoder weights not saved about sted-gaze HOT 4 CLOSED

check this out: NVlabs/few_shot_gaze#19
are you using multi-gpu training?

from sted-gaze.

yes

from sted-gaze.

I managed to fix it by adjusting the load_model function in util.py to remove the "module" word from the keys, you'll need to do something similar for the decoder and discriminator:

if next(iter(checkpoint['encoder'].keys())).startswith('module.'):
    checkpoint['encoder'] = dict([(k[7:], v) for k, v in checkpoint['encoder'].items()])

from sted-gaze.

Hey,
the provided model checkpoint is trained with single gpu, I just tested eval.py with both single and multi gpu training, and i had no problem loading the problem with either case.
Maybe if someone else meets the same problem as you, they can consider to remove 'module' as you did.

from sted-gaze.