facebookresearch / facebookresearch/sam2
path inconsistencies between filesystem and hydra
- Dominant language
- Jupyter Notebook
- Stars
- 19.9k
- Forks
- 2.5k
- PR merge metrics
- No merged PRs in 30d
Description
Great work, but config paths are more complex than they ought to be.
* you have 2 different way to define paths, one for the pth the other for the configs. Cloning and running does not by default work ( probably unless `cwd` is something specific). Hydra may be elegant for some specific pipelines but definitely not for me.
What id like? To use this great project without the hydra mess.
I made it work for me by clearing the default sam2 hydra initialization
```python
import os
from hydra.core.global_hydra import GlobalHydra
from hydra import initialize, compose
from sam2.build_sam import build_sam2 # hydra paths get set on class init
# to use ones own paths clear hydra
GlobalHydra.instance().clear()
sam2_checkpoint = os.path.join(checkpoint_path, "sam2.1_hiera_large.pt") # OK
config_file = os.path.join(config_path, "sam2.1_hiera_l.yaml") # <- this won't work unless one re initializes hydra
# instead do this and overwrite the hydra paths
config_file = "sam2.1_hiera_l.yaml"
config_path = "../some/path/relative/to/cwd"
with initialize(version_base=None, config_path=config_path):
sam2_model = build_sam2(config_file, sam2_checkpoint, device=device, apply_postprocessing=False)
```
also see this https://github.com/facebookresearch/hydra/issues/3060
Contributor guide
Assessment
This issue has not been assessed yet.