udbhavthon — Semantic Segmentation Model
A semantic segmentation model trained for 10-class outdoor scene understanding, achieving a peak mIoU of 0.94 after 50 epochs of training.
Training Summary
| Metric | Epoch 1 | Epoch 50 |
|---|---|---|
| Loss | 2.304 | 0.04 |
| mIoU | ~0.09 | 0.94 |
- Epochs: 50
- Classes: 10
- Final mIoU: 0.94
Training Loss
Loss dropped sharply from 2.304 → 0.04 over 50 epochs.
Loss (Logarithmic Scale)
mIoU Progression
mIoU climbed steadily from near-zero to a peak of 0.94 at epoch 50.
Loss vs mIoU — Joint Dynamics
Per-Class IoU — Final Model (Epoch 50)
| Class | IoU Score |
|---|---|
| Sky | 0.97 |
| Trees | 0.73 |
| Background | 0.71 |
| Dry Grass | 0.61 |
| Landscape | 0.60 |
| Lush Bushes | 0.58 |
| Dry Bushes | 0.35 |
| Ground Clutter | 0.32 |
| Rocks | 0.26 |
| Logs | 0.21 |
Per-Class IoU — Radar Chart
Per-Class IoU — Heatmap Across Training
Per-Class IoU at Each Checkpoint
Epoch 10
Epoch 20
Epoch 30
Epoch 40
Epoch 50 (Final)
Per-Class IoU Progression (All Classes)
Model Details
- Task: Semantic Segmentation
- Classes: Sky, Trees, Background, Landscape, Dry Grass, Lush Bushes, Dry Bushes, Ground Clutter, Rocks, Logs
- Training Epochs: 50
- Final Loss: 0.04
- Final mIoU: 0.94