Initialization of random hash in the neural networks may cause poor accuracy

CapPow / HerbASAP

An effort to automate image post processing for natural history collections

GNU General Public License v3.0

8 stars 3 forks source link

Initialization of random hash in the neural networks may cause poor accuracy #17

Closed dakilaledesma closed 5 years ago

dakilaledesma commented 5 years ago

Most likely, we will need to generate a good seed for startup to maintain consistent results. I believe the rare times that the NN predicts awfully is just through poor luck with the random hash generation.

dakilaledesma commented 5 years ago

False alarm - I was using the high performance discriminator on the wildly color shifted images.