Elucidation/Tensorflow-Cat-Dog-Classifier

Tutorial on using Tensorflow to train a cat/dog classifier, as well as building your own dataset to train on

Tensorflow Cats vs Dogs classifier

Ipython notebook new Mobilenet model

Using TF Keras to transfer learn + fine-tune a MobileNetV2 model on our own dataset of dogs/cats.

We get around 93% accuracy on the test dataset.

Ipython notebook training testing model

Old model

We build our own dataset from existing flicker images of cats and dogs, and then train a tensorflow neural network to classify cats and dogs.

Run using ./cats_vs_dogs.py

Currently a single layer NN, no successful learning yet

usage: cats_vs_dogs.py [-h] [--cat_dir CAT_DIR] [--dog_dir DOG_DIR]
                       [--num_steps NUM_STEPS]

optional arguments:
  -h, --help            show this help message and exit
  --cat_dir CAT_DIR     Directory for storing input cat images
  --dog_dir DOG_DIR     Directory for storing input dog images
  --num_steps NUM_STEPS
                        Number of steps to train model

Building a dataset

resize_images.py contains a script to resize all passed in images into 64x64 grayscale pngs named ####.png monotonically increasing in the specified output folder

Usage: resize_images.py [options] image1 [image2 ...]

Options:
  -h, --help            show this help message and exit
  -o OUTPUT_FOLDER, --output_folder=OUTPUT_FOLDER
                        Output folder to save resized images to
  -n MAX_N, --max_number=MAX_N
                        Maximum number of images to process
  -d, --dryrun          Do a dry run (no processing/saving)