Repository navigation
Jm/hotfix - #3
Conversation
…L-free postprocessors
…etain Medium/Easy AP as secondary metrics
… and abs_rel/RMSE as secondary metrics
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 59abe08f9e
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7580bfc22f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Please focus this review on the decoded YOLOv10 and pose postprocessing changes in 69331f0. Verify the documented ordered class-probability, selected-confidence, and box triplet for single-class face models; confirm multi-class outputs retain the existing class-wise dual top-k semantics; check invalid decoded values fail loudly; and verify compact pose extraction does not bypass non-end-to-end export behavior. Please ignore the already-resolved earlier threads unless the new implementation regresses them. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 69331f042d
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Please focus on the decoded-output validation changes in 5fe3e5a. Verify that all selected-confidence values are range-checked, that every above-threshold multi-class decoded box is validated before top-k truncation, and that compact decoded pose visibility logits are rejected when non-finite before sigmoid. Please also check that standard raw-output paths and legitimate decode-enabled YOLOv10 and pose outputs remain compatible. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5fe3e5aa0c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
@codex review Please focus on f3209e5. Verify that decoded YOLOv10 triplets preserve the standard fixed (B, 300, 6) tensor contract whenever e2e is disabled, while end-to-end behavior remains unchanged. Also verify that DFL-free decoded pose visibility values are checked before sigmoid and that raw output paths remain compatible. |
|
Codex Review: Didn't find any major issues. Hooray! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
No description provided.