JBrightmanAI/yolo26n-sem

---
license: agpl-3.0

pipeline_tag: image-segmentation

library_name: ultralytics

base_model: "Ultralytics/YOLO26"

tags:
  - semantic-segmentation
  - aerial-imagery
  - drone
  - uavid
  - yolo26
  - ultralytics
  - pytorch
  - computer-vision
datasets:
  - dronefreak/UAVid-2020
metrics:
  - miou
  - pixel-accuracy
---

# YOLO26n-sem Finetuned on UAVid

**License:** AGPL-3.0  
**Framework:** Ultralytics  
**Dataset:** UAVid  
**mIoU:** 58.17%  
**Status:** Trained

Fine-tuned YOLO26n semantic segmentation model for aerial UAV imagery using the UAVid benchmark dataset.

This model is part of the **UAVid Semantic Segmentation Model Zoo**, a collection of CABiNet and YOLO26 models trained and evaluated under a common pipeline for aerial semantic segmentation.

---

## Performance

| Metric              | Score           |
| ------------------- | --------------- |
| mIoU                | 58.17           |
| Pixel Accuracy      | 82.31           |
| Parameters (M)      | 1.63            |
| FLOPs (GFLOPs @ 1024px) | 11.4        |

---

## UAVid Model Zoo

| Rank | Model                           | mIoU (%) | Pixel Acc (%) | Params (M) | FLOPs (GFLOPs) |
| ---- | ------------------------------- | -------- | ------------- | ---------- | -------------- |
| 1    | CABiNet (MobileNetV3-Large)     | 68.60    | 87.31         | 9.17       | 54.8           |
| 2    | CABiNet (MobileNetV3-Small)     | 66.84    | 86.45         | 5.36       | 44.1           |
| 3    | YOLO26x-sem                     | 64.41    | 85.82         | 40.16      | 430.9          |
| 4    | YOLO26l-sem                     | 63.28    | 84.75         | 17.87      | 192.4          |
| 5    | YOLO26m-sem                     | 61.98    | 84.41         | 14.32      | 152.3          |
| 6    | YOLO26s-sem                     | 61.69    | 84.27         | 6.50       | 44.4           |
| 7    | YOLO26n-sem (this model)        | 58.17    | 82.31         | 1.63       | 11.4           |

---

## Per‑Class IoU (%)

| Class        | CABiNet (Large) | CABiNet (Small) | YOLO26x | YOLO26l | YOLO26m | YOLO26s | YOLO26n |
| ------------ | --------------- | --------------- | ------- | ------- | ------- | ------- | ------- |
| Clutter      | 69.37           | 67.96           | 67.34   | 65.63   | 64.63   | 63.80   | 61.46   |
| Building     | 87.77           | 86.59           | 87.37   | 85.90   | 86.10   | 85.01   | 82.29   |
| Road         | 81.62           | 80.94           | 79.82   | 78.87   | 78.61   | 78.14   | 75.25   |
| Static Car   | 59.30           | 55.69           | 51.33   | 54.48   | 44.74   | 47.39   | 41.08   |
| Tree         | 81.23           | 80.09           | 78.19   | 76.68   | 76.44   | 76.80   | 74.06   |
| Vegetation   | 65.88           | 64.12           | 63.30   | 59.97   | 60.07   | 60.25   | 55.65   |
| Human        | 30.04           | 27.32           | 21.09   | 19.34   | 20.41   | 18.88   | 15.82   |
| Moving Car   | 73.62           | 71.99           | 66.81   | 65.38   | 64.88   | 63.30   | 59.73   |

---

## Evaluation Visualizations

- **Per‑Class IoU Bar Chart** – `iou_bar_chart.png`  
- **Normalised Confusion Matrix** – `confusion_matrix_normalized.png`  
- **Training Loss Curves** – `results.png`

---

## Dataset

UAVid is a high‑resolution UAV semantic segmentation benchmark of urban street scenes, captured from oblique aerial viewpoints along street‑side flight paths.

**Classes:**
- Clutter
- Building
- Road
- Static Car
- Tree
- Vegetation
- Human
- Moving Car

---

## Usage

### Install Dependencies

```bash
pip install ultralytics

Load Model

Download the model weights file (best.pt) from the repository and load it with the Ultralytics YOLO library:

from ultralytics import YOLO

model = YOLO("best.pt")

Run Inference

results = model.predict(source="image.png", task="semantic", imgsz=1024)
mask = results[0].semantic_mask.cpu().numpy().data  # (H, W) class‑ID map

---

Training Configuration

Setting	Value
Epochs	500
Image size	1024
Batch size	8
Dataset	UAVid (images + masks format)
Framework	Ultralytics YOLO
cls_pw	0.5 (class weight)

---

Training Framework

This model was trained using the CABiNet repository, which provides a unified training and evaluation pipeline for both its own real‑time segmentation networks and Ultralytics YOLO26‑sem models. The pipeline includes shared UAVid dataset tooling and consistent benchmarking across all models. (The repository is publicly available; see the original author’s page for details.)

---

Known Limitations

Performance may degrade in:

• Very small or thin objects (e.g. pedestrians, moving cars at high altitude)
• Heavy occlusion under tree canopy
• Motion blur on moving vehicles
• High input resolutions – UAVid source images are 3840×2160 / 4096×2160; the pipeline evaluates at reduced size (1024px).

---

Citation

If you use this model, please cite the following works (the relevant DOIs and arXiv identifiers are available in the original publications):

• UAVid dataset paper (ISPRS Journal of Photogrammetry and Remote Sensing, 2020)
• CABiNet paper (ICRA 2021)
• Real‑time Semantic Segmentation with CABiNet (ISPRS Journal of Photogrammetry and Remote Sensing, 2021)
• Ultralytics YOLO26 paper (arXiv preprint, 2026)
• CABiNet semantic segmentation benchmarking on UAVid (software repository, 2026)