Cats vs Dogs (Kaggle -Deep Learning)

Problem Statement

Asirra (Animal Species Image Recognition for Restricting Access) requires identifying Dogs and Cats from given images. The current literature suggests machine classifiers can score above 80% accuracy on this task. They need more advance techniques improve accuracy

Techniques

Fine-tune Vgg16 model to solve Cats vs Dogs classification problem.

Tools and Libraries

• Vgg16 model
• Keras
• Scikit learn

Results

Result file is available inside /submission/ directory

• 0.10 score on Public leader board on Kaggle
• top 50% score in Kaggle competition.

Difficulties

• Setup GPU on AWS (as this is my first deep learning project)
• Keep data in proper structure so vgg16 and keras model can read it.

Future Scope

Apply similar techniques to other object detection application/competition like Right Whale Recognition, Galaxy Zoo - The Galaxy Challenge etc.

Reference

• http://course.fast.ai/
• https://www.kaggle.com/c/dogs-vs-cats