TankSense SegFormer-B2
Semantic segmentation model for M1A2 Abrams tank zone detection.
Model Details
- Architecture: SegFormer-B2 (MiT-B2 backbone)
- Task: Semantic segmentation — 8 tank zones + background
- Input: RGB image (512×512)
- Output: Per-pixel class map (9 classes)
- Val mIoU: 0.6578
Classes
| ID | Zone | IoU |
|---|---|---|
| 0 | Background | 0.993 |
| 1 | Hull01 | 0.719 |
| 2 | Hull02 | 0.760 |
| 3 | Turret01 | 0.674 |
| 4 | Turret02 | 0.705 |
| 5 | Track | 0.440 |
| 6 | Wheel | 0.835 |
| 7 | Equipment | 0.428 |
| 8 | Glass | 0.367 |
Training
- Synthetic dataset generated in Unreal Engine 5.5
- 1000 scenes with domain randomisation
- 50 epochs, SegFormer-B2 pretrained on ImageNet
- Downloads last month
- 129
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support