I'm relatively new to ROS2 and looking for suggestions on the best practices for processing batches on image data within a ROS2 node. I need to run predictions with computer vision models for streaming images. The models that I am using perform significantly better, if I run predictions on a batch of images. In my use case, it is acceptable to have a slight delay while accumulating a batch of images, to increase overall efficiency and throughput. I have a working publisher and subscriber, that simply calls predict on a single image within the callback of the subscriber. This works, but I believe i could achieve much better performance processing in batch mode.