chengxuxin/extreme-parkour

Depth image processing.

Opened this issue · 1 comments

Hello!
Thank you for your great open source work. I am currently attempting to implement your work on a real robot. I have observed that the depth images captured by the RealSense camera are quite noisy.

I would be gratefull if you could tell me how you processed image in deployment on real robot. Have you used any filtering(like hole filling filter or temporal filter)? In the paper you write "We preprocess the image by cropping dead pixels from the left hand side and downsampling to 58 × 87" Is this the entirety of your image processing steps?

I appreciate your guidance on this matter.

Have you deployed the policy into the robot?