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
I resolved the AcquireFood issue by doing a couple of things:
Manually add food_spanet_all_rgb_wall_ckpt_best.pth to the expected directory
Map the weights to cpu as follows: ckpt = torch.load(ckpt_file, map_location=torch.device('cpu')). Since I'm fairly sure I used something like pip install cv2 or equivalent, CUDA is not enabled by default. I'm guessing this was done on lovelace, so we should probably add explicit instructions to set that up on other production machines.
Install Pillow version 9.5.0: pip install Pillow==9.5.0. In later versions, you will get AttributeError: module 'PIL.Image' has no attribute 'ANTIALIAS', so we should probably add this to requirements.txt
The default context adapter is SPANet, and by default it looks for this checkpoint. However, it expects a certain repository structure, plus this checkpoint has to be downloaded.
We should add a convenience script that auto-configures the repository structure and downloads this checkpoint.
The text was updated successfully, but these errors were encountered: