The goal of that lab is to create CNN that solves Oregon wildlife Classification task
Pre-requisites:
- TensorFlow 2.x environment
Steps to reproduce results:
- Clone the repository:
git clone git@github.com:AlexanderSoroka/CNN-oregon-wildlife-classifier.git
- Download Oregon Wildlife from kaggle to archive.zip
- unpack dataset
unzip archive.zip
- change current directory to the folder with unpacked dataset
- Generate TFRecords with build_image_data.py script:
python build_image_data.py --input <dataset root path> --output <tf output path>
Validate that total size of generated tfrecord files is close ot original dataset size
- Run train.py to train pre-defined CNN:
python train.py --train '<dataset root path>/train*'
- Modify model and have fun