Closed geoffwoollard closed 5 years ago
See https://github.com/geoffwoollard/ece1512_project/commit/3377459febdf6552827fefed1f0acd070b779c60
This takes 10s of minutes (~2min/epoch x 10 epochs) to fit 1000 examples cropped to 128*128
Maybe if the minibatch size is smaller (ie 100 or 200) it will speed up super-linearly.
Is not fitting (ie training is failing) by epoch 6/10!
Fixed bug and getting 100% accuracy. See https://github.com/geoffwoollard/ece1512_project/commit/33a71be2c60e0c5a866cbf73b1bebbc98a080dd1
write code for architecture in the Deep Consensus paper (Table 1 in doi:10.1107/S2052252518014392)
They used keras, but their code is not released as far as I can tell.