Dog's breed classifer
Machine Learning Capstone Project - Udacity
The goal of this project is to explore the machine learning approach to classifer images, in this case, dog's breed. The architecture of Neural Network to achieve this it's called Convolutional Neural Network (CNN).
You can find the orignal repository from Udacity https://github.com/udacity/deep-learning-v2-pytorch.git, directory `deep-learning-v2-pytorch/project-dog-classification`
The final algorithm should load an image as input and classify as following
- In case of a dog, predict the dog's breed;
- In case of a human face, predict a dog's breed which that human face resembled it;
- In case of neither dog or human, it should provide an error
The steps to achieve this goal
- Datasets
- Download
- Loading
- Pre-processing
- Human face detection
- OpenCV haar cascade to match a human face
- Dog's detection
- Pre-trained network like ResNet50 and VGG16 with 1000 classes, providing great accuracy
- Dog's breed classifier
- Train a model from scratch
- Use transfer learning to use the already trained models as a base for our model
Tools used on this project
- Jupyter notebook / AWS Sagemaker Notebook, Jupyter Notebook is an open-source web application that allows you to create and share documents that contain live code, equations, visualizations and narrative text. It was used from data exploration to final algorithm tests
- PyTorch, open source machine learning library based on the Torch library. It was used to developer the CNN models to predict the dog's breed
- OpenCV, open source computer vision and machine learning software library. It was used to identify the human face
- Matplotlib, plotting library which produces publication quality figures. It was used to show images and plot graphs
- NumPy, library which adds support for large, multi-dimensional arrays and matrices, along with a large collection of high-level mathematical functions to operate on these arrays.
Devices used for testing
- AWS Sagemaker notebooks (small instance and a big one with GPU to speed up the training)
- Notebook i7, 16GB RAM, Nvidia 1070
Web app
On the folder web is available on this repository a small web project which currently is deployed on AWS EC2.
It receives the image, process it loads the trained model and predicts the outcome base on the algorithm defined above.
- The first screen, to input the image
- Result screen with a dog image input
- Result screen with a non dog or human image input
- Result screen with a human face and predicting the dog's breed to that face
