Classifies different species of monkeys, with deep convolutional neural network. For only 30 epoch I achieve 68% accuracy with a custom designed network architecture.
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 we've managed to get a whooping 96.63% validation accuracy. See the transfer.ipynb file.
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: