Dogs vs Cats

Predict whether a given image is of a cat or a dog with 99.7% accuracy.

Kaggle: https://www.kaggle.com/c/dogs-vs-cats-redux-kernels-edition

Partially based on Fast.ai course.

Intro

• TODO: The goals
• TODO: Though process and methods in creating the project
• TODO: Some bullet points with interesting observations
• TODO: Any interesting charts or diagrams

Model info

Final prediction is made from an ensemble of Xception, ResNet50, Inception-ResNetV2.

Training is done using bottlenecks on a data augmented [x6] training set (138000 299x299x3 pictures).

Precomputed bottleneck features are feeded into 2 hidden dense layers [x2048] and 1 output softmax layer.

Size

Training set size = 2.5GB. Model size = 1GB per model (saved precomputed weights).

Speed

Time to precompute bottlenecks: 2hr per model on a GTX 770.

Time to train: 5sec/epoch with 10000 batch size.

Error rates

Ensemble reaches 99.7% accuracy on the validation set (only 6/2000 incorrect).

0.03893 leaderboard score (15th place from 1,314 teams).

Requirements

• keras==2.0.9
• tensorflow==1.2.0
• pandas
• seaborn
• sklearn

Data

Download the data

pip install kaggle-cli
kg config -g -u <username> -p <password> -c dogs-vs-cats-redux-kernels-edition
kg download
7z x *.zip

I split the data into 23000 train and 2000 validation sets.

File structure

dogs-vs-cats-redux
├── data ─── dogscats
│             ├── train
│             │    ├── cats
│             │    └── dogs
│             ├── valid
│             │    ├── cats
│             │    └── dogs
│             ├── test
│             │    └── unknown
│             ├── models
│             └── results
├── submissions
│   └── submission1.csv
├── dogs_cats_redux_ensemble.ipynb
└── README.md

Usage

dogs_cats_redux_ensemble.ipynb for final model

xception_example.py - example of Xception prediction

resnet50_example.py - example of ResNet50 prediction

inception_resnet_v2_example.py - example of Inception-ResNet-V2 prediction