theroyakash / MonkeyNet

Classifies different species of monkeys, with deep custom made convolutional neural network.

MonkeyNet

Classifies different species of monkeys, with deep convolutional neural network. For only 30 epoch I achieve 68% accuracy with a custom designed network architecture.

Dataset

Let's talk about how you can download the data, first you can download the data from here>.

or if you are using google colab or any Jupyter Notebook environment you can run the following command in the cell:

import os
os.environ['KAGGLE_USERNAME'] = "theroyakash"      # Change to your username
os.environ['KAGGLE_KEY'] = "################CONFIDENTIAL################"

You can find your kaggle key from account settings by downloading a JSON file. Now once this set, download the data using the following command in a cell.

!kaggle datasets download -d slothkong/10-monkey-species

Or you can also use the terminal. ZSH or Bash will do. eff the windows. Please don't use windows for your DNN training.

Using transfer learning from InceptionV3

Using transfer learning from InceptionV3 we've managed to get a whooping 96.63% validation accuracy. See the transfer.ipynb file.

Accuracy

Loss

Architecture

Here is the model architecture. It's 23 layer deep neural network. The data input is 200, 200, 3 color channel. The data flows through the network like this: