Orange Puppy Project

Image classifier: Identifies black-backed jackals in trail camera footage.

Acknowledgements

Knowing very little about Machine Learning, this was truly just a fun little thing to do that I cannot claim any superior knowledge over. Saying that, this would not be possible without the various tutorials I followed, and most importantly, the availability of good data. To that I owe the LILA datasets all my gratitude.

Now if only I could get a farmer to actually use this...

Requirements

This project was built on a specific version of Python and Keras. If you wish to run this without any edits, you will have to match the specific Python version.

Python 3.6.*

• install pyenv and download python version 3.6.<any>
• set your python version in the project directory pyenv local 3.6.*

azcopy

• download azcopy from here
• extract 'azcopy' to the project root

Python packages

• numpy pip install numpy
• cv2 pip install opencv-python
• tqdm pip install tqdm
• keras pip install keras==2.3.0
• specific version required pip install 'h5py==2.10.0' --force-reinstall

Images

• run python scripts/download_dataset.py
• if completed correctly, continue with the next step. Otherwise repeat.
• run python scripts/move.py

Modelling & Classification

Parse into correct format

• run python scripts/process.py

Create the machine learning model

• run python scripts/model.py. Make sure to input a name for the model when prompted.

Evaluating pictures

• By default, images in 'test_set' will be evaluated.
• Add any images to this directory.
• run python script/evaluate.py