search

wenbowen123 / iros20-6d-pose-tracking

[IROS 2020] se(3)-TrackNet: Data-driven 6D Pose Tracking by Calibrating Image Residuals in Synthetic Domains
Other
384 stars 66 forks source link

Results on YCBInEOAT dataset. #39

Closed JiChun-Wang closed 2 years ago

JiChun-Wang commented 2 years ago

Hi, bowen, sorry to bother you again. When I run predit.py to evaluate the nine video sequences of YCBInEOAT dataset. All sequences except for the 'bleach_hard_00_03_chaitanya' one perform well. I just use your provided pre-trained weights and the translation normalizer and translation normalizer are set to 0.03m and 30 degree respectively. The following are several visulization results: 0000001 0000050 0000100 0000150 0000323 Hope for your assistance.

wenbowen123 commented 2 years ago

This doesn't look like it's set up correctly. Can you make sure this is set to the correct dataset_info.yml? It should be pointing to some file under the training data you downloaded from https://archive.cs.rutgers.edu/archive/a/2020/pracsys/Bowen/iros2020/YCBInEOAT_weights.zip https://github.com/wenbowen123/iros20-6d-pose-tracking/blob/1c28f02b93aa282a706181d1b28e9a56a9511d3c/predict.py#L630

JiChun-Wang commented 2 years ago

Actually, the other eight video sequences can perform well. I think the necessary setup includes the camera intrinsic, the original image size, the target resolution and the type of renderer and if that's true, it should be set up correctly.

wenbowen123 commented 2 years ago

I just verified from my side and didn't encounter this problem. From your image it looks like it lost tracking very early in the video. The scene so far in the video is not challenging, so it's still likely there is something wrong. Notice the dataset_info.yml here is different from the one used for YCB-Video data.

JiChun-Wang commented 2 years ago

Thanks for your reply, I will try again.