To run this code, do the following:
- clone this repository
- download the dataset and unzip it in the root directory
unzip task.zip
- make installation scripts executable
chmod +x *.sh
- run and installation script for anaconda
./install_conda.sh
or for python virtual envinronment
./install_venv.sh
- for quick running of notebooks in console, change the name of the notebook and obtain script
./run_jupyter.sh
Here is a notebook with exploratory data analysis (EDA.ipynb) and 3 notebooks with different architechtures used to classify images - ResNet18, ResNet34 and ResNet50.
The best result (Accuracy: 92.90%, Loss: 0.0002) after training for 50 epochs has been achieved by ResNet34 model.
All the weights from training can be downloaded from here: https://drive.google.com/open?id=1NAjkxxULFiMJ-7E3n5deUhWWsA5K8c12