search

nv-nguyen / cnos

[ICCV 2023 R6D] PyTorch implementation of CNOS: A Strong Baseline for CAD-based Novel Object Segmentation based on Segmenting Anything and DINOv2
MIT License
191 stars 21 forks source link

Any standard for scaling CAD models? #10

Closed saishruthibalaji closed 1 year ago

saishruthibalaji commented 1 year ago

I have some custom CAD models which have dimensions originally in the order of 0.1m. However, they don't appear at all in the rendered images. So I scaled the mesh to 1000x using blender and then rendered. My rendered images are attached below: some of the regions are getting cut off. (This is mesh_001.ply in this link). Is there some standard in scaling the meshes so that I get perfect renders?

templates

In turn, I'm also not getting any segmented results in the below test RGB Image:

vis

nv-nguyen commented 1 year ago

This issue has been fixed in the most recent commit with this line. Here is what I get for the visualization of templates:

templates

Please note that you can also scale the mesh directly with our code by changing this value radius (distance to the camera) to bigger values, no need to use Blender.

I don't have your RGB image so I made a screenshot of the gray image above and run CNOS. Here is its result with the default hyper-parameters and num_max_dets=20. I think the results can be improved by changing the hyper-parameters and use your input RGB instead of gray image:

vis

saishruthibalaji commented 1 year ago

Thank you! This was very helpful.