/unet_semantic_segmentation

Primary LanguagePython

Semantic Segmentation using UNET

We use the U-Net architechture to do semantic segmentation on Helen Dataset

Architecture

alt text

Dataset

Helen dataset consists 2000 train images, masks and 100 test images, masks with 11 classes.

The dataset has been resized to 256,256 for both images and segmentation masks.

alt text alt text

alt textalt text

Model predictions on test set

alt text alt_text

Uses

  • To retrain
python3 train.py
  • To get class wise f1 scores
python3   path/to/f1_score/    path/to/test/labels    path/to/test/preds  path/to/labels_names.txt