Code fully commented. Credits: Medium
Dataset: Carvana Structure: ./carvana/<train/, train_masks/, train_masks.csv, test/, metadata.csv>
Requisites:
- Python 3.8
- CUDA 11.2
- Pytorch
- CV2 4.5.1
- numpy 1.19.5
- pandas 1.2.3
- torch 1.7.1
- sklearn 0.24.1
- albumentations 0.5.2
Possibile modifications:
- lines 70 - 116: dataset, folders, ...
- lines 220 - 240: trainer parameters
- lines 329 - end: validation set (with groundtruth mask) vs single image from test folder (with original image)