Apply Model

Deep Learning > Apply Model

Applies a trained Deep Learning model to the selected image. The output depends on the model type:

Segmentation models (SegNet/U-Net): Output can be a probability map of any of the trained layers, or a layer map. A probability map indicates each pixel’s likelihood to belong to that layer. A layer map represents the most likely layer for each pixel.

YOLO models: Detected objects displayed as bounding boxes overlaid on the image. Output can be box detections of either one or all of the trained classes.

1. Load Model

Load Deep Learning model. These are created and saved from the AI Session Processor. Additionally, these can be traced PyTorch models.

2. Apply To

Choose image to apply model to.

• Color Image: Model is applied to the originally opened color image (only available for color images).

• Current Image: Model is applied to Current Image.

3. Pre-processing Recipe

Recipe used to pre-process the chosen image before applying the model to it. The recipe used during training (if any) is available and labeled as such. It is also available to load a different pre-processing recipe.

4. Processor

Processor to apply model with.

• GPU: Apply model on the GPU (strongly recommended – see Deep Learning System Requirements for more information).

• CPU: Apply model on the CPU.

5. Size Factor

Resize factor prior to applying model. Image will be resized according to this factor, before the model applied, and then results will be resampled back to original size. It is recommended to use the same size factor as was used to train the model.

6. Output

Choose image to display and output.

For Segmentation Models (SegNet/U-Net):

• Probability Maps: One for each of the trained layers. A probability map indicates each pixel’s likelihood to belong to that layer.
• Layer Map: Represents the most likely layer classification for each pixel.

For YOLO Models:

• Detections for Single Layer: One for each of the trained layers. Bounding boxes shown with confidence score for each detection above the Confidence Threshold for the selected layer.
• All Features: Bounding boxes shown with layer labels for each detection above the Confidence Threshold for all layers.
• Confidence Threshold: Filter detections by confidence level. Only visible when a YOLO model is loaded.

*The final output for YOLO models are grayscale images for compatibility with existing recipe workflows. For “single layer” outputs, box intensities reflect their confidence score. For “all features” outputs, box intensities reflect their class. Use a Basic Threshold following the output to select features.

8. Apply Model

Apply Deep Learning model, with current settings, to the chosen image.

Example

Applying a Deep Learning model