See the source code, it uses worker_index to find which part of the image it needs to process.