Annotation Masks

Upload annotation masks that label the structures you want the model to segment.

Annotation masks are labeled versions of your training images where each pixel is assigned a class value. These teach the model what to segment.

File Requirements

• Format: TIFF stack (multi-page TIFF)

• Bit depth: 8-bit or 16-bit (will be converted to 8-bit if needed)

• Dimensions: Must exactly match your raw images (same width, height, and number of slices)

Pixel Values

• 0 = Background (areas not to segment)

• 1, 2, 3... = Object classes (each unique value represents a different class)

Annotation Guidelines

• Be consistent in how you label structures across all slices

• Ensure object boundaries are accurately traced

• Include examples of objects at different sizes, orientations, and positions

• Label all instances of structures you want to detect — unlabeled objects will be treated as background

• The quality of your annotations directly affects model accuracy