I’ve been generally cropping images closely so just the subject is uploaded; sometimes erasing parts of the images so only the subject is left. Would it be better not to crop or to upload two images: cropped one of subject and uncropped showing surroundings? Many plants have leaves intertwined so it is hard to distinguish parts of subject plant in image. This comment raises the question. https://www.inaturalist.org/observations/26271908
I don’t know that erasing parts of the image really helps very much. Cropping down to show mostly the plant is a good idea, but uploading an uncropped version (or at least not cropped as hard) will also help. Generally, for plants it’s a good idea to provide several views – one showing the entire plant, if you can, another showing the leaves, one showing flowers if possible. I find that even trying that, it’s usually the angle I didn’t shoot that’s the one someone wants for the ID.
I rarely manually crop an image, but I usually take multiple pictures of an organism so that I can post both a closeup of the organism (from multiple angles) and a far shot of the environment (when it seems relevant)
The times I do crop are usually to help the inat AI identify which part of the picture I’m focused on.
I agree with @psweet that erasing parts of the image does not really help, I would recommend sticking with a simple square/rectangular crop rather than the more complex shape crop. An initial image with the target plant at image center is ideal. If the plant is well centered, no cropping need be done. As @erininmd noted, a closeup is a good opening shot. If a photo of the environment is included, be sure the plant is in the photo. If the photo is of the broader habitat without the plant, then that should not be included in the set of observation photos as this will have a negative impact on the training of the computer vision for that plant. This splitting out of the habitat photo can be seen on the observation at:
For a discussion on showing habitat, see: https://forum.inaturalist.org/t/suggestion-on-how-to-include-habitat-images/3553
I don’t crop at all and usually upload several pictures to one obs. Whole habitus, several close-ups (foliage, flowers, stem or bark depending on species) and a snap of surroundings - because almost nothing grows by itself (there are, of course, exceptions) but in “society” so to speak. I’m not really a fan of computer vision right now and i think my approach is good for human identification.
i rarely crop because it’s hard with my setup. I do sometimes zoom in which has the same effect on iphone since it’s a ‘digital zoom’. Cropping has its upsides and downsides, in the long term what would be better would be some sort of in-site ability to tag or frame the target species in the observation. That might also help the algorithm. However, it may be difficult or even impossible to develop. In my head i imagine something like the facebook thing to tag individual people.
It would also help human identifiers if the observation were labeled with something more specific than “Unknown”. My background is mostly botanical, and if I see a plant with an inconspicuous insect on it, or fungal spots on the leaves, I’ll probably try to identify the plant (which is not what the uploading naturalist wanted). Both of these examples have happened to me. Something as broad as “Insects” or “Fungi Including Lichens” would have helped.
There are some related topic threads for this issue where you might want to chime in:
This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.