2 ms·

Thanks for the explanation. Reading the paper, it seems it still takes 2-4 hours to train (one scene?). I imagine that's still faster than any manual method.