Interesting. In that case, the 2048 x 2048 limitation is entirely inexplicable to me.
Or perhaps a “use creative commons licensing and the size limit goes away” approach would be appropriate…
1 Like
There are not just monetary costs to data storage, but also environmental ones…
A lot of arguments against cropping mention losing valuable data. But in fact, if I think abot pictures that could have been cropped here, the data that is lost would not be too valuable… like a lot of sky or water or meadow around a far too small object of interest. Nothing is lost in cropping that.
I think nobody says everything needs to be cropped, but a lot can be without losing anything, but at the same time improving chance of IDing
9 Likes
The thing is, suppose you crop everything that could be cropped. iNaturalist is still going to be deleting a lot of useful data by downsizing images. I was, honestly, kind of pissed off when I learned about the image downsizing, since it was after a bunch of time IDing ferns and squinting at barely visible scales… then learning that iNaturalist had the data that would have made that task dramatically easier, and deleted it.
That’s why “crop your images” is not a solution. Sure, sometimes it’s helpful, but it’s not a solution. Downsizing images is always going to create some level of data loss, and cropping does not change that.
3 Likes
By original files are you thinking the raw format? Any real digital photograph can have these problems:
- JPEG Artifacts - Using JPEG format can introduce visible artifacts in the image if compression is high.
- Gamma correction - Camera equipment and display monitor may have different color correction, causing certain colors to pop and could vary by brand, etc.
- Color Filter - The sensor itself has a color filter in front of it, where the arrangement is often not RGB like a regular monitor, so what the sensor sees vs. what’s in a RGB-type file is different.
- CMOS Sensor - The sensor itself has some gaps between each pixel, averages the photons coming into each pixel (the exposure time) and converts the charge accumulated, corresponding to each pixel, into a binary value. There could be some thermal/electrical noise here, changing the value slightly.
So after all of these corrections, why is cropping any worse than the light averaging done by each pixel? It’s just done spatially instead of over time.
4 Likes
I , too , have said this repeatedly.
2 Likes
Could you elaborate - in comparison to what and at what level would inaturalist photo storage become an issue. There are energy issues for both storage operations and retrieval operations, then there is the mining foot print. I would be curious to know the actual impact and comparisons to some other “environmental issues” as I have no idea of what scale were are talking about.
1 Like
I was wondering the same thing. When people are saying that cropping loses valuable data, it doesn’t make sense to me. Maybe they are conflating cropping and resampling or resizing? When I refer to the benefit of cropping it is to make your photo primarily your subject and removing large areas of irrelevant background. When 80% of your photo area is sky, a white sheet, etc., how is removing that problematic? Then the remaining photo area that actually shows my subject isn’t going to be resampled down upon upload to iNat.
When I read and watch videos about nature photography many of the photographers will say things like never crop, zoom with your feet instead, only take photos during the golden hours for good light, and never shoot through window glass. Many have stated that it isn’t worth your time to take poor quality photos. That may be true for taking the highest quality nature photos if it’s your profession. However, it’s way too limiting and not feasible for most of us iNatters. I have to work with the equipment I have and I’m willing to settle for a pretty poor quality but usable photo to document a sighting if it’s the best I can get at the time. If I see a rare tiny bird in the distance through my kitchen window at high noon, I’m going for it! I might then try to slowly open the window or go outside to get closer, but 80% of the time it’s gone by then. That photo is getting cropped big time too!
I do appreciate the desire to not just throw away resolution, but also understand that there is a cost to data storage and bandwidth. Sometimes we don’t have enough forethought and make compromises that we later regret. Some 20+ years ago I spent a lot of time scanning prints to digitize family photos for a family tree project. Back then storage was a lot more expensive. Those ~90ppi scans of small prints looked pretty decent on my CRT monitor, and I had no plans to print them. Now, I’m kicking myself every time I see one of those tiny photos on my computer!
10 Likes
I’m thinking “don’t have iNaturalist automatically run files through highly lossy data conversions”. Keep the original file as uploaded. What happens prior to upload is out of iNaturalist’s hands, in any case.
I think there would be benefits to adding RAW support to iNaturalist, but also much more substantial costs, and presumably a bunch of development headaches that iNaturalist staff would probably not wish to take on. In comparison, I assume that keeping the originally uploaded jpg and leaving the rest of the system unchanged would be minimally invasive and feasible.
My concern is mostly related to resizing rather than cropping. Cropping is not necessarily a problem, but neither is it a solution to data loss through image resizing. Cropping of course could be quite bad, if you crop out the bits of an organism needed for ID.
3 Likes
I’m curious if anyone knows more related to image hosting costs and the link Marina shared: https://www.inaturalist.org/blog/49564-inaturalist-licensed-observation-images-in-the-amazon-open-data-sponsorship-program/
If iNaturalist doesn’t have server / bandwidth costs for creative commons images, what could be the argument for downsizing them?
1 Like
Anecdotal experience, but when reviewing the higher quality photos through the app, it takes a bit longer for them to load, meanwhile the image shown is a lower resolution one that’s pixelated and unhelpful for ID. So maybe not so much a server-side issue, but could be a factor for user-side, with lower bandwidth? This could be an app glitch too, not sure.
2 Likes
I find I need the big computer screen for IDs, so haven’t really worked with / thought about the app in this context.
To the extent there are transfer speed issues to be resolved, I’m inclined to think that the solution would be similar to how most video streaming sources handle resolution–automatically scale the resolution sent to the user based on the connection speed, plus an option for the user to override that automatic scaling. It seems unlikely that this would be more difficult for pictures than for video, but I don’t really have any idea what’s involved with implementation of that kind of functionality. For all I know, it could be either utterly trivial, or very difficult for reasons that I would never think of.
“Downsample the data and discard the original because sometimes you’ll need to send it over a slow connection” seems very unlikely to be one of the good solutions here.
1 Like
I could be wrong, but most modern streaming codecs use techniques used in H.264, where it encodes differences from frame-to-frame. I’m not aware of any image format that allows for a lower-quality image to be shown if only a fraction of the file is downloaded, but that would be neat!
1 Like
There was a web interface tool for this purpose that was popular with herbarium databases a while ago. Can’t remember what it was called, though. I want to say it had something like “sgt” or “pip” in the name.
Of course, the “dumb” way to do it is to just store a few different sizes of each image, and there’s no shortage of websites that use that basic approach, including massive image repositories like flickr. I don’t know if I’ve seen the image-picker use connection speed, but presumably that would be trivial to someone who’s familiar with the connection speed tester widgets (not me).
1 Like
Ah: MrSID. No idea how it works or if it’s at all relevant to the present topic, but for a few years some herbaria wanting to make large numbers of high-resolution images available via a scalable raster web interface used MrSID. I’m guessing it fell out of use as server costs went down and bandwidth speeds went up.
1 Like
Honest question, isn’t the metadata only important if you’re using the app to automatically fill in those fields for you? I thought it was ok to upload images with no relevant metadata if you correctly provide that information (date, location, etc.) when creating the observation on iNat?
Also, a quick note about cropping. In some cases the context matters, such as insects that live on or around specific host plants. For those, I’d add a non-cropped picture showing the plant a bit better, rather than crop most of it.
2 Likes
If you use the web site to batch upload (ie upload multiple images at one time) images that were taken at different locations and dates/times, metadata becomes quite a time saver
3 Likes
I crop (windows, laptop PC, crop/edit exposure with ACDsee) and don’t have the metadata affected.
2 Likes
I was a little surprised to read the suggestions regarding making adjustments during post processing. Particularly so to make color corrections. In my unqualified opinion, if true color representation is important, don’t do it. Why? Far too many variables. When was the last time you had your eyes calibrated? Your camera? Your monitor? It’s all just too subjective.
The statement was made, “Only crop if you know what you are doing. Otherwise, don’t crop.”
I whole-heartedly agree. On the other hand, how do I know? On yet another hand, how do you know if I know what I’m doing? Since I don’t get any feedback, I must be doing something right. Right?
If you’re using Gimp, and you haven’t tried Paintdotnet, you should try it. IMO it’s far superior to Gimp. If I had a nickle for each artistically inclined person I’ve converted, I’d have… a cup of coffee.
I’m old. I babble. I’ll stop now.
3 Likes
It’s common to assume that the image “straight out of the camera” is basically a neutral, accurate representation of reality. In that light, I would agree–when accuracy is important, don’t modify the color afterward.
That viewpoint is closest to being accurate for RAW images, which come close to being unmodified camera sensor output written to a file. However, camera sensors don’t react the same way our eyes and visual systems do. If we view accuracy as similarity between what we perceive when looking at the image and what we see when looking at the organism directly, then, the RAW image is not necessarily any more accurate than anything else. When the color temperature of the light when the image was taken deviates substantially from the color temperature of sunlight (e.g., at dawn, in the shade of trees, in incandescent light or many fluorescent or LED lights), the colors in the RAW image may be very different from both what we would perceive under the same conditions and what the organism would look like in sunlight.
In any case, most people don’t work with RAW images. If you’re using jpegs, there’s already a lot of processing and modification that happens in the camera. You can often find settings to set that in-camera processing to be relatively neutral. The default settings are designed to create images that people will find aesthetically pleasing, not to create images that are objective or accurate. Most of the time, the difference between an image “straight out of the camera” and an image that someone’s doen post-processing work on is that one was modified by automatic processes, the other modified by manual processes. Neither can be assumed to be more neutral or objective than the other. A human trying to create more neutral, accurate images is probably going to do it better than the built-in image processing defaults on most cameras. A human trying to create an artistic effect might produce images much less accurate than the camera defaults.
8 Likes
I just remembered my own color defficiency. 1 in 12 men are color blind. I can’t tell the difference between red and orange - particularly under flourescent light.
3 Likes