Running Inference

Apply your trained model to new images to generate segmentation masks.

Inference is the process of using your trained model to segment new images that weren't part of the training data. Unlike training, inference only involves the forward pass — no learning occurs.

How it works

1. Your input stack is processed one slice at a time

2. Each slice is passed through the model whole — inference does not tile images into patches

3. The model predicts a class for every pixel in the slice

4. The per-slice predictions are stacked back together

5. The final segmentation mask is saved as a TIFF stack

Because each slice is segmented as a whole image, the slice dimensions must be compatible with the model's depth: the smallest dimension has to be between 32 and 4096 pixels and divisible by 2 to the power of the number of layers. The module runs a pre-flight check and warns you before inference if a stack does not meet this requirement.

The output segmentation mask has the same dimensions as your input and contains integer values corresponding to class labels (0 for background, 1+ for detected objects).

Inference is much faster than training since no gradient computation or weight updates are needed.