Oezys project

A HackKošice 2026 project about developing a disease detection software utilizing the teardrop crystal data. Currently implementing an ensemble of various models to test their detection capabilities.

Used models

• Finetuned ResNet18
• Custom CNN
• RandomForest method

Techstack

• Weights and Biases
• Pytorch Lightning
• Albumentations
• timm
• invoke

Instalation

python3 -m venv env
source env/bin/activate
pip install -r requirements.txt

TODO

• Finish Dataloader
• Finish Dataset
• Integrate invoke
• Integrate andri's cooking
• Set ResNet18 for train
• Utilize Weights and Biases
• Wire Data inspector to validate train data
• Setup Model API that will be called by App
• Finish augmentations
• Send to GPU
• Add debug info in vscode
• Rewire all to andri's new data preprocessing
• Start drawing out pipeline
• ML + DL ensemble MoE
• Debug MoE because it is not really functional right now