Tensorflow Retrain

About

This is an example on how to use pre-existing TensorFlow models to retrain on your data. In this example, we use Inception which is an image classifier. You can train it on the existing images of different noodles type and then use images in the test directory to get the predictions.

Setup

• Install TensorFlow
• Put directories with images in categories directory. The name of the directories will be the classifications

Training

python retrain.py \
  --bottleneck_dir=bottlenecks \
  --how_many_training_steps=500 \
  --model_dir=inception \
  --summaries_dir=training_summaries/basic \
  --output_graph=retrained_graph.pb \
  --output_labels=retrained_labels.txt \
  --image_dir=categories

Predicting

python label_image.py <insert_file_path_to_predict_here>

Visualize

See the model and progress in TensorBoard

tensorboard --logdir training_summaries

Demo

• Here is a presentation of this sample in action. Demo starts at 42:00 - (link)[https://www.youtube.com/watch?v=xlxSVWxeZdM]

Credits

• Albert Padin
• Based on Image Retraining Example