Tiling options
The Tiling options section controls how the input raster is split into tiles before running inference. This mechanism allows the processing to be adapted to data characteristics, model constraints, and hardware limitations.
Tile size
The Tile size parameter defines the size of tiles used for processing, expressed in pixels.
- The value is expressed in pixels
- The tile size must be a multiple of 32
- The default value is 512
Tile overlap
The Overlap parameter defines the amount of overlap between adjacent tiles, expressed in pixels.
- An additional indicator displays the corresponding percentage
- Overlap helps reduce edge effects during raster tiling
Padding
The Padding parameter controls the addition of an extra margin around each tile before inference.
- For PyTorch models, the parameter can be freely adjusted
- For YOLO models, padding is fixed to 0
Batch size
The Batch size parameter defines the number of tiles processed simultaneously during inference.
- For PyTorch models, the parameter can be freely adjusted
- For YOLO models, batch size is fixed to 1 when running on CPU