Hello,
Please help me with quantization parameters. I’m trying to convert my custom tensorflow network:
Input - 8 bit, one channel, 112x112 image
Output - x,y coordinates of 15 points, total 30 values in range of [0,1]
Output example:
[[0.4616952 0.16812855 0.49519286 0.30975538 0.33618504 0.36114803
0.67504096 0.34599686 0.25566477 0.58619547 0.7686454 0.6105488
0.14955398 0.7648848 0.7422619 0.82313496 0.51359755 0.6791271
0.38764966 0.82281923 0.6721683 0.83991355 0. 0.
0. 0. 0. 0. 0. 0. ]
What validation values should I use for validation_tf.txt?
It looks like I need to use integers but how float values like 0.4616952 should be transformed?
Should I scale it back to image coordinates like 0.4616952 x 112 (image width) = 52?
Thank you very much