Fix RT-DETR cache for generate_anchors #31671
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
SangbumChoi
approved these changes
Jun 28, 2024
No, I didn't. I tried bfloat16 only for inference 🙂 |
amyeroberts
approved these changes
Jul 1, 2024
| @@ -1656,7 +1656,10 @@ def unfreeze_backbone(self): | |||
| param.requires_grad_(True) | |||
|
|
|||
| @lru_cache(maxsize=32) | |||
| def generate_anchors(self, spatial_shapes=None, grid_size=0.05, dtype=torch.float32, device="cpu"): | |||
| def generate_anchors(self, spatial_shapes=None, grid_size=0.05): | |||
| # We always generate anchors in float32 to preserve original model code | |||
There was a problem hiding this comment.
Let's update this to reflect the true reason: preserving equivalence between dynamic and static anchor inference. We don't really care about whether our code matches the original model's, just that equivalent logic is used
qubvel
commented
Jul 3, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
For RT-DETR model:
generate_anchorsmethod, so even if we go with dynamic anchors generation we only generate it once per image of the same size.Who can review?
cc @SangbumChoi