Load and Label

Preparation

Prior to using this plugin, put the images you want to work with in a project directory as shown below.

📌 Load Panel

After starting the plugin the first step is to load your images and assign labels.

1️⃣ Click the Open image directory... button.
2️⃣ Select the directory that contains your image files.

Drawing Labels

1️⃣ Select Label box layer and draw a label box that is as large or larger than the desired patch size.
2️⃣ Select labels layer and Label objects within the label box.

Sparse vs Dense Labeling

Some algorithms support sparse labeling, which requires less work than dense labeling.

Dense Labeling

Label every object in the image. Any pixels not labeled are implicitly assumed to be background. All pixels essentially have a label (either explicitly labeled objects or implicit background). Works with all frameworks.

Sparse Labeling

Label only some objects and some background regions. Pixels with value 0 are treated as unlabeled and masked out during training (not used). This is faster but not all frameworks support it.

Important Notes: - Value 0 = unlabeled (masked out during training) - Label 1 = background (must label some background explicitly) - Labels 2+ = different object classes (for instance segmentation) or pixel classes (for semantic segmentation). - Internally, the framework may subtract 1 from labels (making unlabeled pixels -1, background 0, etc.)

Framework Support: Not all frameworks support sparse labeling. Check if your chosen framework supports this feature by: - Consulting the framework documentation - Asking on image.sc - Posting on other public forums with details about your problem

Sparse labeling can significantly reduce annotation time, but make sure your framework supports it before relying on this approach.

Save Results

Select Save Results periodically to save the labels you have drawn.

After saving results folders should be generated for different types of deep learning artifacts.

Inspect the labels directory to verify labels you have drawn have been saved.

🔄 Next: Set Validation Labels