Repository navigation
Inclusion of Important Object IDs in test/validation set #35
Description
Activity
In issue #11 @ChonghaoSima mentioned: "we assume the model already 'know' the key objects". Dose this imply that the model should be capable of identifying key objects even amidst varying c-id or locations from the predictions?
Thank you for a good question. We put some effort here to bypass the problem of id-issue.
- The model still needs to identify the key objects.
- If the model identifies the key objects with different id-number in the gt (the c-id in your sentence), we will use the IoU matching of the predicted and the GT 2D coordinate, to find the corrent object in your predicted answer. That being said, the c-id is not a matter here. You could have a look at our evaluation code about this.
Thanks for the comment!
- pinned this issue
on Mar 26, 2024
Hi there, will the test/validation set include important object IDs (the "key_object_infos" dict)? If not, how will questions of subsequent steps (prediction/planning, etc.) be composed with the ID?