Segmentation¶
The segmentation workflow consists of three main steps:
- Boundary Prediction
- Boundary to Superpixels
- Superpixels to Segmentation
Widget: 1. Boundary Predictions¶
Choose one of the build-in PlantSeg models, or one from the BioImage.IO Model Zoo.
Alternatively, you can import your own model by choosing ADD CUSTOM MODEL
from the model selection drop-down menu.
- Mode: Select the mode to run the prediction.
- Model filter: Choose to only show models tagged with `plantseg`.
- PlantSeg model: Select a pretrained model. Current model description: Unet trained on confocal images of Arabidopsis Ovules on 1/2-resolution in XY with BCEDiceLoss.
- Show advanced parameters: Change the patch shape, halo shape, and batch size.
- Device: None
- Mode: Select the mode to run the prediction.
- Model filter: Choose to only show models tagged with `plantseg`.
- BioImage.IO model: Select a model from BioImage.IO model zoo.
- Show advanced parameters: Change the patch shape, halo shape, and batch size.
- Device: None
Widget: 2. Boundary to Superpixels¶
Here, the boundary prediction is turned into superpixels by using distance transform watershed.
None
- Mode: Define if the Watershed will run slice by slice (faster) or on the full volume (slower).
- Boundary threshold: A low value will increase over-segmentation tendency and a large value increase under-segmentation tendency.
- Minimum superpixel size: Minimum superpixel size allowed in voxels.
- Show advanced parameters: Show advanced parameters for the Watershed algorithm.
Widget: 3. Superpixels to Segmentation¶
DT Watershed tends to over-segment the image, therefor an agglomeration algorithm is used in this third step.
- Agglomeration mode: Select which agglomeration algorithm to use.
- Under/Over segmentation factor: A low value will increase under-segmentation tendency and a large value increase over-segmentation tendency.
- Minimum segment size: Minimum segment size allowed in voxels.