Comment on page
Configure Tiled Model
Tiled Model Block adds tiling capabilities to allow small objects to be detected. Tiling splits images into smaller sections for models to process each tile separately. Images are then restitched back together to create inferences for the image as a whole. Image tiling is often used to detect small objects in high resolution imagery.
Click to configure and add a Tiled Model block to your pipeline.
To add a tiled model to your pipeline:
- 1.Navigate to the "Pipeline Blocks" tab.
- 2.Under the Prediction section, click on the Tiled Model option to configure this processing block.
- 3.Next, select the Model Name and Model Version of the model you want to add from the list. Model Details will populate. Note: The model selected does not have to already be trained with tiling to be used in the Tiled Model block.
- 4.Under Classes, select which classes (labels) you want to return predictions for.
- 5.Using the slider, select the desired Confidence Threshold (if available). This value defaults to 60%. Predictions must meet this confidence score in order to be returned.
- 6.Under Tile Settings, select whether you want to use pre-defined tiling or custom tiling options.
Select the tile size (400x400px, 600x600px, or 800x800px) and pre-defined options will be used for tile overlap depending on the size selected.
Set the following values for custom tiling:
- Tile Size - Height and width of each tile. Accepted range is 200px - 800px with a default of 800px.
- Custom Tile Overlap (Stride) - Specify the tile overlap or stride. This value is a percentage of the tile dimension and defaults to 50%.
When finished, scroll down and click "Add to Pipeline" to add the block to your pipeline. This will bring you back to the "Pipeline Blocks" tab. From there you can add additional blocks and drag and drop to rearrange around other blocks as needed.