amirhosseinh77 / UNet-AerialSegmentation

A PyTorch implementation of U-Net for aerial imagery semantic segmentation.

U-Net for Semantic Segmentation on Unbalanced Aerial Imagery

Note: this repository is still developing but the colab notebook is complete for training and evaluations.

Read the article at Towards data science

Kaggle Dataset

UNet-AerialSegmentation
       		   ├── dataloader.py
       		   ├── losses.py
      		   ├── model.py
      		   ├── train.py
       		   └── inference.py

Training

!python train.py --num_epochs 2 --batch 2 --loss focalloss