You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
During initial construction of the Eyeballer model, the Imagenet weights are downloaded from the internet. This is performed even when using the pretrained weights from Eyeballer. The need for Imagenet to be downloaded prevents Eyeballer from being usable offline.
What should the expected behavior be
Imagenet weights should only be downloaded when training a new model and not when used with the "predict" flag.
This can be resolved by setting the weights during initial construction to "None" instead of "Imagenet" in the model.py file.
Steps to Reproduce
Delete the cache of Imagenet. Disable internet connectivity and perform a prediction with Eyeballer using the pretrained weights.
Additional context
Eyeballer was used on an internal assessment by someone unfamiliar with the code base. This bug report was requested by them.
The text was updated successfully, but these errors were encountered:
Description of Bug
During initial construction of the Eyeballer model, the Imagenet weights are downloaded from the internet. This is performed even when using the pretrained weights from Eyeballer. The need for Imagenet to be downloaded prevents Eyeballer from being usable offline.
What should the expected behavior be
Imagenet weights should only be downloaded when training a new model and not when used with the "predict" flag.
This can be resolved by setting the weights during initial construction to "None" instead of "Imagenet" in the
model.py
file.Steps to Reproduce
Delete the cache of Imagenet. Disable internet connectivity and perform a prediction with Eyeballer using the pretrained weights.
Additional context
Eyeballer was used on an internal assessment by someone unfamiliar with the code base. This bug report was requested by them.
The text was updated successfully, but these errors were encountered: