mamonraab / resnet-tensorflow

Modular Residual Networks implemented in TensorFlow. Easily change hyperparameters in a few lines.

Residual Networks in TensorFlow

Residual Network in TensorFlow

This entire code is implemented in pure TensorFlow and I have made it simple to run with different settings.

Simple Instructions

• Running Training and Evaluation

  • python main.py
    • If you want to modify any parameters, you can use for example python main.py --n_epoch==10
      • The default runs on CIFAR-10 dataset and this configuration is made for that.
      • n_epoch: number of epochs
        • Default 10
      • n_batch: batch size
        • Default 64
      • n_img_row: dimension of image (row)
        • Default 32
      • n_img_col: dimension of image (col)
        • Default 32
      • n_img_channels: number of channels
        • Default 3
      • n_classes: number of classes
        • Default 10
      • lr: learning rate (momentum optimizer)
        • Default 0.1
      • n_resid_units: number of residual units
        • Default 5
      • lr_schedule: number of epoch for the learning rate to decrease by lr_factor
        • Default 60
        • This multiplies the LR every 60 epochs by lr_factor.
      • lr_factor: the factor for reducing LR
        • Default 0.1.

• Running TensorBoard

  • Training logs
    • tensorboard --logdir=train_log
  • Evaluation logs
    • tensorboard --logdir=eval_log
  • You can use any path you want.
    • If you encountered a permission denied error, you can easily solve it by changing the directory to tmp/train_log.
    • I experienced this while running on Amazon AWS and it was solved with this fix.

Credits

• The original model was obtained from the official repository by TensorFlow.
• The paper on Residual Networks on arXiv.org.

Dependencies

• To simplify the code, I read the CIFAR dataset using TensorLayer.
  • Simply run sudo pip install tensorlayer and you are good to go.
• TensorFlow v0.12
  • If you would like to run this code in a few minutes on Amazon AWS, just use the open-source AMI TFAMI.v3.

License

MIT