I’m wanting to do some experiments with different neural nets. I’m interested in creating a neural net that can take an image input of a plant/animal/fungi, and identifying what species it is. My thought was to make a tiered neural net system. First NN would identify the kingdom it belongs in (plant, animal, or fungi), then based on the result, it would pass the image to a second NN that’s trained for identifying the phylum, and so on. I’m hypothesizing that this tiered system will have a lower error rate than a single NN, for identifying species.
I want to avoid having redundant image sets. For example, a folder for animals, then a sub-folder for arthropods, then a sub-folder for ants, or something to this effect. I would like to have the parent folders empty, and the final sub-folder to contain the images. Is there a way for fastai to work with this type of file structure?