Parameters for fine-tuning

[noahstier]

Can @zmurez, or anyone else, suggest a configuration for training / fine-tuning Atlas that can learn acceptably well for experimentation using only a single GPU? I wonder if zmurez discovered such a configuration during development.

I will try decreasing VOXEL_DIM_TRAIN, NUM_FRAMES_TRAIN, and RESNETS.DEPTH. However, the convergence time is so long that a hint here could save a lot of time.

Thanks