Introduction
A modern GPU can process thousands of images per second. But before an image reaches the GPU, something has to read it from disk, decode the JPEG, resize and augment it, and copy it to GPU memory. If that work is slower than the GPU, the GPU sits idle, waiting. You've paid for a race car and you're stuck behind a tractor.
An efficient data pipeline makes sure the next batch is always ready before the GPU needs it. Think of a restaurant kitchen: the chef (GPU) is fastest when prep cooks (data workers) have already washed, chopped and laid out the ingredients for the next dish while the current one is cooking.