search

huangjh-pub / multibody-sync

[CVPR'21 Oral] MultiBodySync: Multi-Body Segmentation and Motion Estimation via 3D Scan Synchronization
59 stars 5 forks source link

Test data #6

Closed estherlxd closed 2 years ago

estherlxd commented 3 years ago

Hi @heiwang1997,

Thanks for your work! While I tried to use my own data to test I found if I follow the 'Usage' part in the readme, pass in a (1, K, N, 3) tensor, only contains the xyz value for the points, I could get the result. But in 'Training and Test' part in the readme, even the test data contains three parts: pc, segm and trans. I tried replacing the pc part in sample dataset with my own point clouds value, looks like when I use different npz file with different segm and trans to run the test.py, the result will also be different. And I also found that the paper also mentions The test dataset also contains manual annotations of the object segmentation mask and rigid absolute transformations. Could you please let me know if the segm and trans are needed for using test.py? If yes, what's the reason to make this test method different from the 'Usage' part? Thank you!

Best, Esther

heiwang1997 commented 3 years ago

Hi @estherlxd, thank you for your question! Sorry about the late response.

That is indeed interesting because we do not use any segmentation or transformation information during the test. Could you please, for example, send me the npz file you generated so that I can take a close look?

Best, Jiahui

heiwang1997 commented 2 years ago

Closing due to inactivity. Feel free to re-open if the issue still exists.