Update README.md

# HMN-2f: Age Prediction Model

HMN-2f is a convolutional neural network trained to estimate a person's age from facial images.
It represents the final model in the HMN research line developed by Humanlike before the project was discontinued.

The model is based on **EfficientNet-B3** and was trained on a combination of the UTKFace and FG-NET datasets. HMN-2f focuses on improving generalization across diverse age groups through dataset expansion and strong augmentation techniques.

## Model Details

- **Architecture:** EfficientNet-B3
- **Framework:** PyTorch
- **Input resolution:** 300×300
- **Training datasets:**
- UTKFace (~23,000 images)
- FG-NET (~3,000 images)
- **Task:** Age estimation (regression)

## Performance

Benchmark results measured on the **APPA-Real dataset**:

| Metric | Score |
|------|------|
| Mean Absolute Error (MAE) | **6.835 years** |

Additional controlled webcam tests demonstrated significantly lower error in favorable conditions.

## Intended Use

This model is intended for:

- Research into age estimation methods
- Academic experimentation
- Educational purposes
- Computer vision experimentation

## Limitations

Age estimation from facial images is inherently difficult and error-prone.

Known limitations include:

- Reduced accuracy for extreme ages
- Demographic imbalance inherited from training datasets
- Sensitivity to lighting, pose, and occlusion
- Performance degradation on low-quality images

Benchmark scores on difficult datasets may differ significantly from controlled testing conditions.

## Ethical Considerations

Facial analysis technology carries significant ethical risks.
Potential misuse includes:

- Surveillance systems
- Discriminatory profiling
- Privacy-invasive biometric analysis

The HMN research line was discontinued partly due to these concerns.

Users of this model should carefully consider ethical implications and comply with all applicable laws and regulations.

## License

This model is released under the **BigScience OpenRAIL-M license**, which restricts harmful uses of the model.

See the LICENSE file for full terms.

## Citation

If you use this model in research, please cite:

Humanlike (2026). *HMN-2f Age Prediction Model.*

## Acknowledgements

This model was trained using datasets including:

- UTKFace
- FG-NET Aging Database

The EfficientNet architecture was originally developed by Google Research.

README.md

@@ -1,3 +1,16 @@
----
-license: bigscience-openrail-m
----
+---
+license: bigscience-openrail-m
+datasets:
+- nu-delta/utkface
+metrics:
+- mae
+type: null
+base_model:
+- google/efficientnet-b3
+tags:
+- computer-vision
+- age-estimation
+- facial-analysis
+- pytorch
+- efficientnet
+---