still harbor
Hello here I am after about a month with functioning convolutional neural network, with multiple layers and even optimizers.
It's still on developing stage and needs more testing to do, but wanted to share the progress.
Feedbacks are appreciated
https://github.com/saeol0415/CNN
latent quiver
Nice! What sort of score are you hitting on Mnist, if you've tested with that?
still harbor
Well when trained with 1000 images, im getting about 90/100
Should I train with more data?
Its taking me half a minute for single epoch when I train with 10000 images
Or should I increase number of convolutional layer?
still harbor
Today I read this paper: https://arxiv.org/pdf/1502.03167
and made some changes
added batch normalization layer
got average loss of 0.000133028 and 95% accuracy with 100 epoch with 10000 images
||After training for literally an hour🥲 ||
CNN-from-scratch (1.4k+ lines)
safe idol
How to run this
Specify it