Fast-SCNN Floor Segmentation
Lightweight Fast-SCNN model for indoor floor segmentation.
- Task: Semantic Segmentation
- Class: Floor
- Input: RGB
512×512
- Pixel Accuracy: 94.9%
- mIoU: 87.8%
Model Formats
- PyTorch:
.pth
- ONNX:
.onnx
- OpenVINO:
.xml + .bin
Performance
OpenVINO FP32: 26.02 FPS / 38.44 ms per frame on Intel Core 7 240H.
Limitations
Designed for indoor floor segmentation. Performance may degrade with
unseen environments, lighting changes, shadows, furniture, and floor-like
surfaces.
The model detects floor only and does not perform obstacle detection.
