Modular Residual Networks implemented in TensorFlow. Easily change hyperparameters in a few lines.
MIT License
This entire code is implemented in pure TensorFlow and I have made it simple to run with different settings.
Running Training and Evaluation
python main.py
python main.py --n_epoch==10
n_epoch
: number of epochs
10
n_batch
: batch size
64
n_img_row
: dimension of image (row)
32
n_img_col
: dimension of image (col)
32
n_img_channels
: number of channels
3
n_classes
: number of classes
10
lr
: learning rate (momentum optimizer)
0.1
n_resid_units
: number of residual units
5
lr_schedule
: number of epoch for the learning rate to decrease by lr_factor
60
lr_factor
.lr_factor
: the factor for reducing LR
0.1
.Running TensorBoard
tensorboard --logdir=train_log
tensorboard --logdir=eval_log
permission denied
error, you can easily solve it by changing the directory to tmp/train_log
.sudo pip install tensorlayer
and you are good to go.MIT