A PyTorch implementation of U-Net architecture to perform pixel-level semantic segmentation on the Football Scene Dataset. Includes custom mask preprocessing, robust data augmentation with Albumentations, and fine-tuning with Class Weighted CrossEntropyLoss to resolve data imbalance for small object detection.
emseekim/UNet-Semantic-Segmentation-Football
Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|