Closed derneuere closed 3 years ago
I can see the model in the image once the container has booted up but I don't see when in the boot up process grabbed. Do you happen to know when it occurs exactly? If not I can try working through the startup process.
I also don't why it is loaded at bootup. AFAIK it is only used here : https://github.com/LibrePhotos/librephotos/blob/dev/api/places365/places365.py
It should be downloaded when it starts scanning, but that does not seem to be the case.
At startup the backend of librephotos downloads a pytorch model. This model is 250MB big and is download at start every time. This of course is inefficient, because it is a dependency that doesn't change. It should be integrated in the docker steps and the windersnet.py should be rewritten to not download the models but load them from the model integrated from the docker steps.
https://github.com/LibrePhotos/librephotos/blob/6c5796b818b4d1373f20d81d1e2cba641cd70624/wideresnet.py#L9-L15