search

yinyunie / ScenePriors

Implementation of CVPR'23: Learning 3D Scene Priors with 2D Supervision
https://yinyunie.github.io/sceneprior-page/
MIT License
73 stars 6 forks source link

Question on number of objects #7

Closed DianCh closed 1 year ago

DianCh commented 1 year ago

Hi! I have a few questions on the number of objects to render and train: it looks like the maximum number of objects for ScanNet is set to 7 - what if the scene contains less than 7 objects? How does the hungarian matching and batched mesh rendering work? On the contrary, when the scene contains more than 7 objects, does it mean the extra objects are simply ignored?

Thank you very much!

yinyunie commented 1 year ago

Hi,

If I remember, I set max_n_obj = 53. Could you please share with me where I set it to 7? Thanks.

Best, Yinyu

DianCh commented 1 year ago
Screen Shot 2023-08-30 at 9 45 39 AM

from models/ours/modules/network.py

Here for example max_len got the value of 7 in the batch during training (which isn't set fixed, I misunderstood), which I realized that won't result in extra objects being ignored; but how does it work for extra meshes - like when there are 7 meshes for the second sample with 5 objects?

Thanks!

yinyunie commented 1 year ago

Hi,

max_len is the maximum number of objects in a batch. For example, if there are two scenes in a batch - Scene A has 5 objects, and Scene B has 7 objects, max_len will be 7. I also loaded an object mask for each scene, so Scene A has a 7-d binary vector mask with the first 5 entries occupied, while the others are zeros.