---
license: apache-2.0
datasets:
- rwitz/golf-cart-yolo-data
---
# GolfCart YOLO11s

**Model name:** `golfcart-yolo11s`


A compact YOLO11s detection model fine-tuned to identify golf carts in aerial and ground-level images. Optimized for fast inference on edge devices while maintaining strong precision across varying lighting and backgrounds.

---

## Model description

This model is a lightweight YOLO11s variant trained to **detect golf carts** in diverse environments — fairways, parking lots, garages, and drone captures. It balances speed and accuracy for real-time applications like fleet monitoring, course management, safety analytics, and automated observation systems.

* **Architecture:** YOLO11s (small) — one-stage single-shot detector optimized for real-time inference.
* **Input size:** variable; recommended 640×640 or 512×512.
* **Output:** detection mask and confidence score for the class `golf_cart`.

---

## Visual examples

Below are example inputs and model outputs:

![Golf cart — example 1 ](golfcart_01.png)

![Golf cart — example 2 ](golfcart_02.png)

![Golf cart — example 3 ](golfcart_03.png)

---

## Intended uses

**Primary (appropriate):**

* Monitoring golf cart activity on courses.
* Fleet usage and movement tracking.
* Safety or proximity alerts on maintenance vehicles.
* Drone-based detection of golf carts in aerial surveys.

**Out-of-scope:**

* Identifying individuals or non-cart objects.
* Use in safety-critical or privacy-sensitive contexts without human oversight.

---

## Training data

* **Total images:** 624
* **Content:** golf carts and mixed scenes (including negative examples with no carts)
* **Augmentations:** horizontal flip, brightness/contrast variation, mosaic augmentation, scaling, rotation.
* **Annotation style:** detection labels (not bounding boxes)

---

## Evaluation

* **Validation/Evaluation set:** 68 images (balanced with and without golf carts)
* **Results:**

  * **mAP@50:** 0.93
  * **mAP@50–95:** 0.649
  * **Precision:** 0.978
  * **Recall:** 0.853

These results indicate strong performance even with a relatively small dataset, especially in high-confidence detection of golf carts across mixed environments.

---

## How to use

**Install:**

```bash
pip install ultralytics
```

**Inference example:**

```python
from ultralytics import YOLO

model = YOLO('best-3.pt')
results = model('test_images/golfcart_drone_01.jpg', imgsz=640)

results[0].show()  # or results[0].save()
```

Supports ONNX, TorchScript, and TensorRT conversion for deployment.

---

## Limitations and failure modes

* Misses can occur when golf carts are small, heavily occluded, or extremely far away in aerial imagery.
* Performance may vary for uncommon cart shapes or regions not represented in the training set.

---

## Biases

The dataset contains mostly North American golf cart models and course designs. Accuracy may dip in unfamiliar regional styles or environments.

---

## Ethical considerations

* Avoid using this model for surveillance involving people.
* Respect local drone and privacy regulations.

---

## Model parameters

* **Weights:** `best-3.pt`
* **Training date:** 2025-10-03 (replace with actual)
---

## How to cite

If you use this model, please cite with the model name and link to its Hugging Face card.

---

## Contact

Maintainer: Ryan Witzman — `ryan@rwitz.com`