Closed daisily727 closed 4 years ago
Hi, I used [0,255] range
Hi, I used [0,255] range
Thanks for your prompt response. One more question, in the gradient loss computation, did you use multi-scale (e.g. 4 scales in previous papers)?
Hi, sorry for the delay. Predictions are upsampled to full resolution before loss computation, and about the gradient term yes, it takes relevant predictions (e.g. from scale 0,1 and 2) and apply a weight to each one (lower the scale, lower the weight)
As midas output range is from 0 to a few thousand, I tried to use this output as supervision but found the loss cannot decrease at all. If I scale the output to 0 to 1, the training will converge to all zero output very quickly. Can I know what range you used in your training? thx