search

NVlabs / FoundationPose

[CVPR 2024 Highlight] FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects
https://nvlabs.github.io/FoundationPose/
Other
955 stars 99 forks source link

Some issues with the custom dataset #92

Closed tobottyx closed 2 weeks ago

tobottyx commented 3 weeks ago

Thank you for your excellent work.I followed the guidelines from #44 to collect data, but encountered some issues. First, there is a significant discrepancy between the size of the bounding boxes and the actual size of the objects. Secondly, the pose accuracy is not precise, and when I manually adjust the objects, the pose is unstable. I have double-checked all my files, including the format of the depth maps. I used the obj model from ycb-v, and aligned the color and depth images. I would like to thank the auther and @savidini for your help.

https://github.com/NVlabs/FoundationPose/assets/150102559/af69df68-45d7-4131-ac5f-ff866ebca240

Here are some of my results. Below are some .ply files from debugging. I hope the author can take a look for me. Thank you very much. https://drive.google.com/file/d/1cNCCbVohiq45m27u3c_SrCDTo2bU9YC5/view

tobottyx commented 3 weeks ago

0000000000

wenbowen123 commented 3 weeks ago

your depth image is wrong. The point cloud is in the wrong scale and skewed.

tobottyx commented 2 weeks ago

Thank you very much for your answer, I will modify and try again