A herbarium cannot afford to train a leaf-net from empty weights. It starts from a network already fitted on a broad plant-image corpus, then spends a much smaller labeled set on its own sheets. What alternative should the tester summarize?

Starting from a network already fitted on a broad plant-image corpus, then spending a much smaller labeled set on the herbarium's own sheets, is a pretrained model plus fine-tuning. That is cheaper and lower-data than training from empty weights.