Normalise image and keypoints values
Closed this issue · 1 comments
nik1806 commented
Images need to be normalized based backbone network. It is necessary for good learning.
Optionally, values of key points can be normalized in the range of (-1, 1).
Usually, normalization is performed to avoid overshooting of the lose value due to anomaly in a few datapoints. It also speeds up the convergence to minimum.
A transform method is needed for both the purpose. It can be done in 2 steps.
nik1806 commented
While training the loss is overshooting enormously. Need this class asap.