You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have tried to use the iPhone config and training loops forever in the validation stage.
model>nerf.py>Graph>render
line 325
whileray.isnan().any(): # TODO: weird bug, ray becomes NaN arbitrarily if batch_size>1, not deterministic reproduciblecenter,ray=camera.get_center_and_ray(opt,pose,intr=intr) # [B,HW,3]
Hi @kurtjcu, there's a weird bug (as mentioned in the comment) that I haven't been able to resolve here, which could (sometimes) make ray contain NaN values if the batch size were larger than 1. If you happen to find the root cause, please help submit a PR and I would be happy to merge!
What is the code or option change to make the batch_size=1 Doing it in the options did not work.
Just modify model/barf.py get_pose() function, add eps to center_aligned = (center-sim3.t0)/sim3.s0@sim3.R*sim3.s1+sim3.t1, that is center_aligned = (center-sim3.t0)/(sim3.s0 + 1e-7)@sim3.R*sim3.s1+sim3.t1
I have tried to use the iPhone config and training loops forever in the validation stage.
model>nerf.py>Graph>render
line 325
my conda env
The text was updated successfully, but these errors were encountered: