Kvasir-SEG Custom U-Net

This is a custom U-Net built from scratch for polyp segmentation in endoscopic images.

Model Details

  • Architecture: 4-layer U-Net with BatchNorm and Dropout2d.
  • Loss Function: Focal + Dice Loss (So the Network sees the small polyps)
  • Resolution: Trained on 256x256 images.
  • Data Augmentations: Spatial augmentations and color variation

Performance

  • Validation IoU: 0.463803
Downloads last month
46
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support