Download data from: https://www.kaggle.com/c/dogs-vs-cats/data

To convert jpegs into TFRecords (one-time run)
$python convert_to_records.py

Evaluation can be run independently of training.
Training script saves checkpoint files, which evaluation script can read.
(Must run training for sufficient time to generate a checkpoint file. Default is 1 epoch)

To train the model:
$python train_cnn.py

To evaluate the model on validation set:
$python eval_cnn.py 'validation'

To evaluate the model on test set:
$python eval_cnn.py 'test'

To visualize training data on TensorBoard:
$tensorboard --logdir=tmp/train_data/

To visualize evaluation data on TensorBoard:
$tensorboard --logdir=tmp/eval_data/

Currently getting 91% accuracy on validation set.

Jupyter Notebooks were made initally for learning and debugging purpose.
Not part of the main project.

Code highly inspired from tensorflow tutorials.