Closed asheeshr closed 7 years ago
Also, is there a reason behind not performing histogram equalization on the dataset prior to training?
I did not experiment with lots of different normalization strategies. If you use a more data-driven normalization, you should perhaps increase the window size to compensate. Tell me how this works for you ;)
How would the performance change if using per-image mean rather than dataset-mean for training? Would there be significant degradation in the results?