LiteHand-YOLO

LiteHand-YOLO is a lightweight, attention-enhanced YOLOv8n-based model for single-class hand detection.

Source code: https://github.com/ZahraAlipour703/LiteHand-YOLO
Task: Hand detection
Model configuration: litehand-yolov8n.yaml
Checkpoint: litehand_yolov8n.pt
Status: Research and development model

Model overview

LiteHand-YOLO modifies the YOLOv8n architecture with attention modules. The project aims to reduce model complexity while maintaining hand-detection performance.

The model uses a single detection class (nc: 1). Refer to the source repository for the architecture configuration and implementation.

Reported evaluation results

The source repository reports the following results:

Metric	LiteHand-YOLO
Parameters	2.06M
GFLOPs	6.4
mAP@50	99.26%
mAP@50–95	91.42%
Precision	97.82%
Recall	98.26%

The repository also reports a CPU benchmark of 8.83 FPS and 113.22 ms latency on an Intel Core i7-6700 system with 8 GB RAM, using its stated 640 × 480, 300-frame test.

These values are reported results from the project, not an independent reproduction. Results can vary with the dataset, hardware, software versions, and evaluation procedure. Consult the source repository for the benchmark implementation and comparison details.

Training data

Before making this model public, complete this section.

Dataset name: [Enter the exact dataset used to train this checkpoint]
Dataset source: [Provide its official URL or source]
Dataset version and evaluation split: [Enter the actual version and split]
Dataset license and redistribution terms: [Verify and document the applicable terms]

Do not publish the checkpoint publicly until you have checked that the dataset's terms permit the intended release.

Intended use

This model is intended for research, educational work, and experimental evaluation of hand detection.

It has not been established as suitable for safety-critical applications or as a validated medical device. Performance may be affected by camera position, lighting, occlusion, background, and differences between the training and deployment data.

Installation and inference

Important: This checkpoint uses custom module definitions from the LiteHand-YOLO source project. Installing the standard ultralytics package alone may not be sufficient to load it.

Clone the source repository:

git clone https://github.com/ZahraAlipour703/LiteHand-YOLO.git
cd LiteHand-YOLO

Create and activate a Python environment, then install the project's local Ultralytics implementation:

Windows PowerShell
```
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -e .\yolov8-enhanced-structure
```
Download litehand_yolov8n.pt from the Files and versions tab of this Hugging Face repository and place it in the project's weights directory, using the filename litehand_yolov8n.pt.

Run the webcam demo:

python .\demo\realtime.py --model .\weights\litehand_yolov8n.pt --source 0 --device cpu

See the GitHub README for additional image, video, and benchmark instructions.

Limitations

The reported evaluation results should be independently reproduced before being used to support strong performance claims.
Detection performance may change across datasets and deployment conditions.
The checkpoint depends on the custom implementation and compatibility modules in the linked source repository.

License

The project repository uses the GNU Affero General Public License v3.0 (AGPL-3.0). Review the license and all applicable dataset and upstream-component terms before redistributing the checkpoint.

Downloads last month: -; Downloads are not tracked for this model. How to track