Once again about computer vision. Uploading tons of plants from Russia, I noticed that it is hard (or almost impossible) to train the system to understand plants from Eurasia if there are closely related (and look similar to) North American relatives. This is mostly due to the great number of North American pics which were used to calibrate the network. So, dozens and hundreds of sedge pics (genus Carex) from Russia are already uploaded, but still these species are not listed in top-10 suggestions where North American sedge species predominate. The same is true for some other groups.
For instance, Russian hybrid Aronia mitschurinii (or x Sorbaronia mitschurinii) sometimes is listed as a suggestion among other aronias but only with a “Seen nearby” tag. How many pics of Aronia mitschurinii do we need to see one day “Visually similar”?
So, I want to suggest more intensive use of geographical filters at least on continental or macroregional scale. Another suggestion which could play a vital role in precise geographical filtering is to include all GBIF-data for “Seen nearby” suggestions. This could be very helpful in larger countries sparsely sampled by iNat users or for tropical regions where the diversity is very high.