| --- |
| tags: |
| - detectron2 |
| - pointrend |
| - semantic-segmentation |
| - medical-imaging |
| - thyroid |
| library_name: detectron2 |
| --- |
| |
| # Thyroid PointRend Semantic Segmentation |
|
|
| This model is a PointRend-based semantic segmentation checkpoint fine-tuned for thyroid lesion mask prediction. |
|
|
| ## Base Framework |
|
|
| - Detectron2 PointRend |
| - Custom semantic segmentation fine-tuning workflow in `projects/PointRend/inference_thyroid.ipynb` |
|
|
| ## Test Metrics (ThyroidXL test split) |
|
|
| - num_images: 2094 |
| - mean_iou: 0.742992 |
| - mean_dice: 0.835550 |
| - micro_iou: 0.777697 |
| - micro_dice: 0.874949 |
| |
| ## Inference Settings Used For Export |
| |
| - subdivision_steps: 7 |
| - subdivision_num_points: 224 * 224 |
|
|
| ## Files |
|
|
| - `model_final.pth`: fine-tuned checkpoint |
| - `test_summary_iou_dice.txt`: evaluation summary |
| - `metrics.json`: training/evaluation log from output directory |
|
|
| ## Notes |
|
|
| - This repo is private as requested. |
| - Data and labels are not included. |
|
|
