🎯 BreastBox Detection (YOLOv8s+)

The Precision Tool for Breast Detection in Stable Diffusion

BreastBox is a custom-trained object detection model (YOLOv8s+) designed specifically to solve common detection failures in ADetailer. While default models often struggle with side views, overlapping clothing, or dynamic poses, BreastBox provides a tighter, more anatomically correct bounding box.

🔥 Why Use BreastBox?

• ✅ High Precision: Manually curated dataset, moving from 200 to over 650 high-quality labels.

• ✅ Challenging Angles: Expertly handles "lying down", "side-boob", "from behind", and partially obscured views.

• ✅ Anatomically Correct: Designed for tighter masks that follow the natural shape of the body.

• ✅ Style Agnostic: Optimized for 3DCG (DAZ/Poser), but generalizes well to realistic and 2D/Anime styles.

📊 Version History

• v1.0 (Experimental): Initial release (~200 samples). Best for standard poses.

• v2.0: Major data upgrade (~400 samples). Better versatility for breast sizes and camera angles.

• v3.0+ (Latest): Focus on obscured breasts (hands, objects) and smaller chest sizes (~700+ samples).

🚀 Quick Start (ADetailer)

1. Download the

   .pt

   file.

1. Move it to:

   stable-diffusion-webui\models\ADetailer

2. Select

   BreastBox

   from the ADetailer dropdown in WebUI.

1. Recommended Settings:

   • Detection Confidence:

     0.3

     to

     0.5

   • Mask Merge Mode:

     Merge

🛠️ The Power Behind the Training: YOLO Training GUI

I trained and refined this model using YOLO Training GUI, a standalone tool I developed to make YOLO training visual and simple for everyone.

If you are training your own models and want to see your results like I do, check it out: 👉 YOLO Training GUI on GitHub

Why you'll love it:

• Live Monitor: Watch detections happen in real-time as the model trains.

• Side-by-Side Comparison: Generates the dual-view comparison images.

• No Code Required: A full GUI to manage your datasets, hyperparameters, and exports.

💬 Feedback & Community

This is a passion project! If you encounter edge cases or have suggestions:

• Share your Feedback in the comments.

• Post your Example Images (it helps me refine the dataset!).

• If you like the model or the training tool, a Star on GitHub or a review here is much appreciated! ⭐