GoogLeNet/InceptionNet for CIFAR10/100
Pytorch implementation from scratch.
$ git clone https://github.com/axeloh/GoogLeNet.git
$ cd GoogLeNet
$ python train.py
Example of train with other than default params:
$ python train.py --dataset cifar100 --epochs 100 --bs 128 --lr 5e-3 --gpu False --modelname mymodel --save_every 10 --lr_scheduler False
(GPU strongly recommended!)
Model saved in models/
, loss and accuracy plot for train and validation set saved in output/
.
Model trained on CIFAR100
Without data augmentation and learning rate scheduler:
Loss | Accuracy |
---|---|
With data augmentation and with a step learning rate scheduler:
Loss | Accuracy |
---|---|