A classifier using transfer learning on the inception model

This is a response to Siraj's challenge of the week.

This is a classifier made with basic TensorFlow, using Transfer Learning from the Inception model. Data is taken from this Kaggle competition, as recommended by Siraj.

Pipeline

JPG images --> Inception --> 2048-dimensional vector --> Fully connected layer --> Prediction

How to use

Create bottlenecks

First, create bottleneck files for use in the training of the one-layer neural network.

python producebottlenecks.py

This will download the inception network, run your images in the train directory through the inception network, then output the images as bottleneck text files in the bottlenecks folder.

Train on the bottleneck files

python classifier.py

This will train a single layer neural network on the bottleneck files. The trained graph will be saved as savedgraph.pb. Also, you can view TensorBoard using tensorboard --logdir=log on the command line.

Test on any image

python testclassify.py --image_dir=path/to/image

Notes

Only 1000 images of cats and 1000 images of dogs are in the train folder. I did not use all the images as that would take quite a lot of time to produce bottleneck files for. 1000 images of each category are more than enough to produce a very high accuracy.

The code for downloading the inception model and producing the bottleneck text files was taken from the TensorFlow github -> examples -> image_retraining. However, I only took the relevant portions, hence making the code much shorter and easier to understand.