Skip to content

[fix] Fix input size selection in decode-only inference - #41

Open
sarhakor wants to merge 1 commit into
InterDigitalInc:mainfrom
sarhakor:fctmv13_bugfix_sct_decodeonly
Open

sarhakor wants to merge 1 commit into
InterDigitalInc:mainfrom
sarhakor:fctmv13_bugfix_sct_decodeonly

Conversation

@sarhakor

Copy link
Copy Markdown
Contributor
  • Issue: In the CompressAI-Vision version bound to FCTM v13.0, decode-only inference overwrites the model input size with the original image dimensions. This causes incorrect bounding box scaling, resulting in negative task performance (MOTA) under SCT conditions.

  • Fix: Updated the condition to skip the gt_inputs branch when vision_model_info is available, ensuring that the correct model input size stored in the bitstream is used.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant