Udacity Deep Learning Nano Degree Doc Classification Project
This project uses Python to create a CNN to classify dog breeds from supplied images.
The first part of the implementation is based on a neural network created from scracth with the following architecture:
- 4 convolutional layers increasing the depth from 3 to 128 with batch normalization and max pooling between each layer. These convolutional layers create feature maps of the images.
- 2 hidden linear layers and a linear output layer to classify the detected dog breeds.
The second part uses a pretrained network to peform the feature maps and the trains a linear layer to classify the dog breeds from the images.
The simplest way is to install Anaconda and start the Jupyter Notebook dog_app.ipyndb. The supplied setup.sh fil contains command to install some of the required resources, such as the training and validation data. The training was run using the following AMI at Amazon:
Deep Learning AMI (Ubuntu) Version 20.0 (ami-0827ddd2d8e38aa56)