search

swtyree / hope-dataset

HOPE-Image and HOPE-Video 6-DOF pose datasets.
57 stars 3 forks source link

Follow your paper to make a new dataset #5

Closed lyltc1 closed 9 months ago

lyltc1 commented 2 years ago

I want to follow your paper or similar paper to make a new datasets, from scanning to take pictures and annotate pose. Can you show me some tools or code please? Thank you!

swtyree commented 2 years ago

Hi @lyltc1, we used a couple of custom tools for annotation. Our primary tool operated by solving for pose using correspondences between the object mesh and the 2.5D RGB-D point cloud. tool_depth_left_sm tool_depth_right_sm

Another tool did the same using PnP and correspondences between the mesh and the image. tool_pnp_sm

We also had another tool for "nudging" the translation/rotation by small amounts for a better visual alignment. tool_preview_2_sm

I'll see if I can clean up the code and attach it here. The code may or may not work out of the box, but it should give you a starting point.

swtyree commented 9 months ago

Hi @lyltc1, I see that I never got that code cleaned up. In case an annotation tool is still useful to you (or someone else reading this issue), take a look at the code that we just released for creating the HANDAL dataset: https://github.com/NVlabs/HANDAL.