Medical Annotation: Corneal Endothelium Cells Masks

This is a specialized LoRA designed to generate binary segmentation masks for corneal endothelial cells. It produces high-contrast, strictly monochrome (black and white) images representing cell boundaries and structures.

Key Features:

• Output: Binary Masks (Black background/White cells or vice versa).

• Usage: Perfect for generating synthetic ground truth data for medical image segmentation tasks (e.g., training U-Net models) or creating procedural biological textures.

• Workflow & ControlNet Integration: This model is highly effective for synthetic data generation when combined with ControlNet.

  • Creating Paired Datasets: You can use this LoRA to generate a binary mask first, and then feed that mask into ControlNet to guide my model "Medical SEM Style: Corneal Cells".

  • Recommended ControlNet Weight: 1.5

  • Result: This workflow produces perfectly aligned (Image, Label) pairs, which are essential for training segmentation networks (like U-Net) without manual annotation.

Recommended LoRA Weight: 1.5 Base Model: SD 1.5

Training Data & Configuration:

• Dataset: 50 manually annotated masks of corneal endothelial photomicrographs.

• Training Strategy: Robust training with 40 repeats per image.

• Total Steps: 1,000 steps.

• Batch Size: 2

• Resolution: 512x512