Yolov8 Animeface

🚀 yolov8_animeface

Anime Face Detection using YOLOv8

✨ Features

• This model is designed for anime face detection, leveraging the power of YOLOv8.
• It has been trained on a manually - annotated dataset of 10,000 images from safebooru.
• The model shows high precision and recall in detecting anime faces.

📦 Installation

No installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

No code examples are provided in the original document, so this section is skipped.

📚 Documentation

Dataset

The dataset was created with 10,000 images manually annotated from safebooru. The split was 70% for training, 20% for validation, and 10% for testing.

Performance

This model is based on yolov8x6. It was trained on the said dataset for 300 epochs at 1280px * 1280px. It took approximately 110 hours to train on an RTX A4000.

On the dataset, the model performs particularly well with the default parameters.

Images  Instances    Box(P        R      mAP50     mAP50 - 95):
1002       1562      0.957      0.924      0.955      0.534
Speed: 1.3ms preprocess, 81.9ms inference, 0.0ms loss, 0.8ms postprocess per image

Although it doesn't provide a high mAP50 - 95, its predictions are always correct on the tested files. The confidence could be higher, but the model is very precise.

Confusion matrix of the model, generated on the test set

PR curve of the model, generated on the test set

Other metrics are available in the ./demo/val sub - folder.

Demo

Manual annotation	yolov8_animeface

Comparison with an existing model

While we can debate the comparison between two models from different generations, it is interesting to compare this model with zymk9's model, based on yolov5. They both have the same purpose and were trained in the same way.

On the same dataset with the same parameters (conf = 0.001 & iou = 0.6), yolov8x6 - animeface produces better metrics than the one on yolov5x.

yolov8 - animeface:

Images  Instances    Box(P        R      mAP50     mAP50 - 95):
1002       1562      0.956      0.919      0.953      0.532
Speed: 1.2ms preprocess, 81.2ms inference, 0.0ms loss, 1.0ms postprocess per image

yolov5 - anime:

Images  Instances    Box(P        R      mAP50     mAP50 - 95):
1003       1566      0.778      0.685      0.633      0.232
Speed: 1.5ms pre - process, 85.5ms inference, 1.4ms NMS per image at shape (32, 3, 1280, 1280)

yolov5 - anime provides better results when images are resized to 640px, but it is still inferior to yolov8 - animeface with the same parameters. Surprisingly, yolov5 is much more confident than yolov8, but it also has many more false positives.

Yolov8_animeface	Yolov5_anime

Conclusion

While this model performs well on the dataset, the metrics may differ from others. A lighter model might be released later.

Based on YOLOv8 by Jocher, G., Chaurasia, A., & Qiu, J. (2023). Ultralytics YOLO (Version 8.0.0) [Computer software]. https://github.com/ultralytics/ultralytics

🔧 Technical Details

The model is based on Ultralytics/YOLOv8. It was trained on a specific dataset with defined parameters and achieved certain performance metrics as described above.

📄 License

The model is released under the AGPL - 3.0 license.