3 ms·
Fascinating. It touches on the discolored and counterfeit parts but doesn't say how they are detected – I assume there was a lot of manual training of the neur
by tomcam 10y ago
Fascinating. It touches on the discolored and counterfeit parts but doesn't say how they are detected – I assume there was a lot of manual training of the neural net?
- jacquesm 10y agoIt's basically bootstrapped. Use the machine to sort a few kilos, pick out the mistakes, update the training set, let it train overnight, rinse, repeat. After many such cycles it is getting pretty good. The best part is when I think it has made a mistake but actually it is right and I'm wrong :)