I have Some conceptual queries regarding the code:

When I put the random_pose = False in line 396 of mpi_rendering.py and pass the same pose as given by the dataloader([flat_w2c_mats_real] line 419 train.py ) to self.mpi() function [line 451 mpi_renderer.py] then after 6 iteration on 8 gpu the error arises .

Error:

pos: tensor([[0.5797, 0.0647, 0.1850]], device='cuda:0')

dir: tensor([[-0.6583, -0.1611, 0.7355]], device='cuda:0')

u: tensor([-1.0399, -1.0401, -1.0398, -1.0396], device='cuda:0')

v: tensor([-0.6758, -0.6710, -0.6656, -0.6603], device='cuda:0')

yaws: [[-0.6179566383361816]]

pitches: [[0.07507812976837158]]

Traceback (most recent call last):
File "/proj/cvl/users/x_fahkh/mn/debug-dmpi/gmpi/core/mpi.py", line 106, in homography
assert torch.min(u) >= -1, f"Ray's U direction goes out of plane at {distance}, min val {torch.min(u)}"
AssertionError: Ray's U direction goes out of plane at tensor([[1.1200]], device='cuda:0'), min val -1.0400704145431519

The question is that how is it possible that extrinsic parameter[flat_w2c_mats_real] of real image to leads out of plane of the rays.