Hello,
This PR addresses an issue we encounter when having to use SuperPoint in a production environment.
Currently, the pre-trained model weights MUST be downloaded from https://github.com/rpautrat/SuperPoint/raw/master/weights/superpoint_v6_from_tf.pth.
This is not compatible with a production env:
You don't necessarily want your model to access internet
GitHub could be down for a moment, thus your endpoint could be down as well
Rémi Pautrat could rename / delete that file, just to watch the world burn 😄
This PR allows the users to specify a weights parameter in the config input (similar to LightGlue), so that the weights are loaded from cache instead of from rpautrat's repo.
Hello, This PR addresses an issue we encounter when having to use SuperPoint in a production environment. Currently, the pre-trained model weights MUST be downloaded from
https://github.com/rpautrat/SuperPoint/raw/master/weights/superpoint_v6_from_tf.pth
.This is not compatible with a production env:
This PR allows the users to specify a
weights
parameter in the config input (similar toLightGlue
), so that the weights are loaded from cache instead of fromrpautrat
's repo.