Nan in loss

[zyzwhdx]

Hi, many thanks for the great work.

But when I was training this model, the training log showed Nan in loss, and this happened even in your uploaded exp folder. I guess maybe it is caused by the smooth loss calculation (pred_distance).

pred_distance = pred_distance[:args.N_rgb], pred_distance_coarse = pred_distance_coarse[:args.N_rgb], pred_distance_smooth = pred_distance[args.N_rgb:]

Do you have any suggestion for this problem? Thank you!

[booker-max]

I also encountered the same problem, args.N_rgb = 4096, and len(pred_distance) = 4096, hence pred_distance_smooth = [], and ultimately caused NAN loss.

[booker-max]

I don't know how to solve it, Do you have any advice?