ylabbe / cosypose

Code for "CosyPose: Consistent multi-view multi-object 6D pose estimation", ECCV 2020.
MIT License
301 stars 89 forks source link

Question about the BOP20 results #45

Closed zmccmzty closed 3 years ago

zmccmzty commented 3 years ago

Hi Yann,

Thanks for sharing the great work!

I have some question about the BOP20 results.

Let's take the result for the LMO dataset as an example: This is the result for scene 2 image 3 I downloaded from this repo: Screenshot from 2021-03-28 10-16-20 This is the example result provided by the BOP website for the same dataset, same scene and same image: Screenshot from 2021-03-28 10-16-33

This is the ground truth of the list of objects appeared in this scene: Screenshot from 2021-03-28 10-16-58

The problem is that in this scene, there are only 8 objects in the scene (obj 1, 5, 6, 8, 9, 10, 11, 12). And in the example result from the BOP website, for each image, it only provides an estimation for each object in the scene. But in the results of CosyPose, there are estimations for objects which are not in the scene (like obj 1, 2, 4, 7, 15), and there could also be multiple estimations for an object (like obj 8, 9 and 12). Besides, obj 10 is missed in this image.

I would like to try some ICP methods based on your prediction result, and I am not sure how should I use your results correctly given the difference between it and the example results provided by the BOP website. Could you please give me some explanation on your results?

Thank you in advance!