Run to run variation between runs with the same input images

[sbhatia31]

Hi,

Thanks a lot for your hard work and amazing contribution. I have been able to reproduce good novel views outputs using the SW, but I noticed, few things as mentioned below:

Run to Run variation at Camera Parameter Estimation step in between runs [Significant differences in the Camera Parameters]
Run to Run variation at MPI generation step even with same camera parameters i.e MPIs generated are different in between the runs
Run to Run variation happens at different resolutions i.e 1080p, 480p, 360p etc
Run to Run variation occurs with different content too, including Fern.

Is this expected? I am able to reproduce at my end, on two different machines.

Looking forward to hear from you soon :) Thanks a lot in advance!!

Thanks,
Sumit

[sbhatia31]

Got run to run fix by Disabling TensorFlow CuDNN Autotuning. Thanks!