Visually retouching images?

General
1

Last week I used editing to get a better taxon photo for this species: http://www.inaturalist.org/observations/182726345.
This editing did not include altering any of the features that would lead to a different identification, but instead just patching up the damage on the shell. My understanding of it is that it’s allowed, because you can completely draw organisms from scratch and get research grade observations of them. Is this the consensus?

Edit: I have removed it as the taxon photo and replaced it with an unedited version ;)

1 Like
2

I crop and adjust white balance, saturation and other image properties a high end camera would do.
I would not reduce noise or retouch.
It should be fine I think, as long as you attach the original.

7 Likes
3

I could be wrong, but from what I heard, its okay to make slight edits to anything surrounding the organism being observed, but not the organism itself.

1 Like
4

I’d suggest not to do this normally. Removing a black spot caused by dirt on the lens, or adjusting white balance and exposure, etc, seem ok. But retouching an image runs the risk of creating characteristics that could mislead about what was actually observed. In this case, you’ve done the right thing by adding the untouched photo and a comment saying what you’ve done, but I wouldn’t suggest doing this regularly.

7 Likes
5

Thank you! But my main concern is that fact that you can draw organisms. Wouldn’t that mean you can “draw” onto a photo of an organism?

I’ve heard to always assume honesty when ID’ing drawings. In that case, we are assuming that the observer drew what they saw. But in this case, I collected a shell, and it broke. Parts of it flaked off. So you’d assume that the edited photo is what the shell looked like before I accidentally damaged it. Right?

6

The aim of drawing an organism is to create a visual representation of the organism as it exists. No representation is perfect — all processes involve processing of photon data in different ways — but the aim is the same: To capture how the organism appears, not how I would like it to appear.

Retouching features of an organism, including “fixing” damaged aspects isn’t trying to represent the organism as it exists. Instead, retouching is trying to create an idealized visual representation of the species, not of the organism itself, much like retouching photos of humans aims to create idealized humans to fulfill an aesthetic, social value. So I do not conclude this to be the same as drawings of the person, since drawings too can be done for different purposes: Accurate and precise depiction, e.g. as one does with anatomical drawings for dissection or as a botanist trying to record the 3D appearance of a flower later to be pressed and preserved. Aesthetic depiction where an artistic license is employed to distort physical reality in order to communicate some form of visual message, e.g. any work of Picasso (at the extreme end) reducing forms to lines and squiggles or Michelangelo drawing more idealized human forms based on real muscular structure, but not necessarily representing real humans.

There’s also data present in those “imperfections”. Scientists might find evidence of a particular disease or a genetic variation present in that individual. Erasing “imperfections” risks erasure of that data.

14 Likes
7

You can do this, but I agree with other commenters that I think this is not recommended, and I would say particularly not for a photo to be used as a taxon photo.

When we edit a photo, we often don’t fully know what we are changing, what details are added/different, or how the photo may be used in the future.

For instance, I know that some people use AI-based denoisers and similar, and I have no problem with these for shots for personal use. I also think that they are probably allowed based on iNat’s guidelines, but I personally am skeptical of their use, simply because it involves creation of detail from a black box method that isn’t understood by the observer.

For instance, assume an AI-based retouching tool has been trained on photos of insects with lots of setae - the AI retoucher is given a photo of an insect and fills in detailed setae as part of the retouching. BUT for some insects, presence or absence of setae is a diagnostic character. In this scenario (which is totally made up, but possible), an unretouched photo without detailed setae visible might be IDed to genus since a key character wasn’t visible. BUT an AI-retouched photo could be IDed to an incorrect species based on the presence of a trait introduced by the AI which was not present. So the consequences of editing are unclear and involve altering data in some ways that can’t easily be replicated or tracked.

In many cases, it might not matter, but in some cases it might.

As an aside, I don’t think drawings are actually a particularly good analogy to use here because they are so rare on iNat. They are allowed because iNat’s goal is to connect people to nature, and this is a way that some people connect. There are definitely concerns about whether people draw what they see (or what they think that they see), especially if a drawing is done after the fact. Of course, a drawing is not going to be used for a taxon photo, so I think the impact is limited.

8 Likes
8

To be a bit more clear about the situation, I wasn’t only fixing the damage on the shell. I previously had the shell when it was undamaged, so I was showing a representation of the organism. I also was using an old photo of the unbroken shell for reference.

You could add the unedited photo with a note saying something like, “The first photo is retouched with AI so it may not be reliable for an ID.”

But thank you for the response. I do agree that there are risks to retouching photos.

The reason I brought up drawings is because what they represent may not be accurate, just like editing an image. You did mention that in your post. But I don’t think it’s that bad of an analogy. I drew the shell but I had both a physical part of the shell and a reference photo.

I’ll change the taxon photo when I get the chance, though :)

1 Like
9

Comparing retouched photos to drawings misses a key point. When someone looks at a drawing it is obvious it is a drawing created by someone. When someone looks at a retouched photo, even if there are notes that say it is a retouched photo, the person might not realize it is a retouched photo.

Also, I would assume you would not want CV training to be affected by a retouched photo.

7 Likes
10

I would keep the issue of “drawings of organisms” and touched-up photos as separate discussions. As a perpetrator of several “observations” based on my own field sketches, I understand all the caveats and skepticism that (appropriately) come with such records. They are judged on their own merits, warts and all. Touching up photos seems to me a fundamentally different process with a goal of making an image more recognizable. I concur with others in this thread that this type of alteration of photographic evidence should be avoided. Let a photo (or a drawing) stand on its own merits, flaws and all.

6 Likes
11

Color balance, contrast, white balance, sharpening, cropping, etc all good, but I would not make changes to the actual content of the image like that. Damage is part of the observation, and sometimes things can be seen in the damaged portions that can assist in observations.

3 Likes
12

Thank you all for your responses! I now realize that a clamshell is very different from a birds, plants, and insects, which probably explains the variety of answers. However, I still think was wrong in editing the photos and I’m changing both the observation and the taxon photo for the species.

I would agree with you for most small organisms like plants, insects, fungi, and more. But I guess in this case, the damage I fixed isn’t covering anything needed for ID, and the “striae” lies mostly unbothered. I have added the original photo though, for comparison.

I would too, and there are other forum topics for that, I believe. In my previous posts for this topic I kept on talking about how “drawing” to fix images is practically the same as sketching an organism from scratch, but I was mistaken, because most drawings on iNat are the field sketches you mentioned, and not the digital drawings I had imagined. Sorry about that.

But, I still am yet to understand the difference between drawing the organism from memory and drawing the organism from memory but I have part of the organism that I unfortunately broke (this is rather specific to clamshells, but still.)
See this illustration I made:

IMG_3461
IMG_34611625×1625 465 KB

The top half shows a drawing I made (from memory.) It shows the identifying features so I guess it could be research grade.

The bottom half shows how I changed the image. Part of the periostracum (the brown layer) was missing, so I drew it in from memory, whilst not changing the features that make it this species.

But looking at it now, I see that the editing was really unnecessary and the original photo is already great. I shouldn’t have changed it in the first place.

I see your point, but if they aren’t reading the description, then how would they be ID’ing it? But then again, in this case, it used to be the taxon photo. As the taxon photo, it wouldn’t show the description when you first look at it.

Wouldn’t sketches affect the CV as well?

I find this more common with plants and other animals, but I do get your point and I have added the original photos.

I do see this now, but I did have an “ideal” specimen (before I slightly damaged it, which is why I did the retouching in the first place.)

Thank you for the well-written response!

Yup, this is really common in bivalve observations in particular. But, these aren’t “slight edits.” Oftentimes, people just remove it completely. This is because the photos are ex situ instead of in situ.

See some of Steven’s bivalve observations: https://www.inaturalist.org/observations?place_id=any&taxon_id=47584&user_id=stefadrian&verifiable=any

He takes great photos and sometimes edits them to have a black background to make ID easier.

1 Like
13

(I have changed the taxon photo to the unedited version of mine. Thank you to the person who had changed it to a different photo earlier, but I feel as if an in situ image would be more representative of what you would see when encountering this organism.)

Edit: I see that you do not want the photo as the taxon photo. I will no longer change it to that then.

14

Definitely. There are even iNat Projects devoted to “damage”, especially on plants.

3 Likes
15

As I interpret the rules all observation photos must be a representation of a specific organism as you observed it, meaning that fixing damage would only be allowed if the damage happened after you saw the organism, and you are restoring the image to what it looked like before being damaged, as that is functionally the same concept as drawing what you saw. However, I usually wouldn’t recommend this as any alteration increases the risk of inaccuracies, and I think it is very important to prominently note that any “fixed” image is altered so people are aware that it has some of the inaccuracy risks of a drawing, and do not think it is as reliable as a normal photo

I would consider any observation where damage that occurred prior to the organism being observed was touched up to be a form of fake or misrepresented observation

Now I want to distinguish touching up damage and similar physical alterations of the organism from other forms of image editing, I do not think denoising, sharpening, contrast adjustments, exposure adjustments, ect are the same, as these are just doing the same things a digital camera already does to the image, and not altering the actual form of the organism is a misleading way, these could be suboptimal if done in a way that resolution or clarity on an identifying feature is reduced, but this is no more misleading than having your camera on the wrong ISO, it’s just lowering the quality, not fabricating anything about the organism (now I do not recommend edits that lower quality, that kind of defeats the point of editing, but it’s not dishonest)

One last point, exposure and contrast changes can alter the color substantially if done to an extreme, and it is important to clearly note if you have to edit something so much that the color is unrealistic

For example here I have a realistic image where noise was reduced, and sharpness and contrast enhanced, I think this is fine, the ant is the exact color as what I saw with my own eyes https://www.inaturalist.org/observations/183571028

While here I took a picture with the ISO so horribly wrong that the amount of editing needed to make the organism show up altered the color substantially, so I disclosed this in the observation notes (and it is obvious that something is off with this image) https://www.inaturalist.org/observations/183023659
But there are situations where more subtle changes can alter the color, so I think it is important to check that the color is still real before you click the button to finalize the edit, but I don’t think this justifies a prohibition on contrast and exposure edits, just don’t go too far and always disclose if you have to do something drastic

3 Likes
16

The CV doesn’t learn distinguishing traits of organisms. It learns visual patterns. Drawings are visually so different than photos that I imagine the CV either treats them as outliers or learns them as a separate pattern for the taxon, much the way it learns to recognize both larval and adult forms of arthropods.

Whereas a retouched photo would share many characteristics with the non-retouched photos in its training set and not be recognized as different from them.

2 Likes
17

Taxon photo history carefully recorded by iNat
https://www.inaturalist.org/taxa/130965/history

1 Like
18

I can see how drawings/sketches are somewhat related to altered photos (though I think there are key differences as well, such as that a drawing is generally obviously not a photo). I think it likely that edited/retouched photos are much more common than drawings though.

I think that sketches could affect the CV if included in the training set, but they are just so infrequent I think. I can probably count on two hands the number of drawings I have seen as observations in my IDing (which would put the frequency at <1/10000), though I really only ID lizards. At this frequency, the CV should be pretty robust to the occasional drawing in the training set. I would also guess that drawings achieve RG (or a CID) less frequently than photo based observations and would be included in the CV training set less frequently than photos, even given their background rate of submission for observations.

The only real issue I could see is if a taxon has a significant number of drawing-based observations like a class project where students are all drawing an uncommon species or something like that. Do others have different experiences with drawings.

1 Like
19

I would guess that individual drawings would reach RG more often because of how much they stand out on the identify page. However, there are definitely more research grade photo observations than drawings.

20

@lj_lamera – I think you handled your photo manipulation well for iNaturalist by saying what you did and by posting the original photo. I hope everyone who changes a photo does those things.

I also think you asked this question because you’re aware of the potential problems. For one thing, those of us who ID quickly may look at the first photo and make and ID if that seems clear, not noticing the rest. (I know I should look a little closer but, well, sometimes I don’t.) Also, important details may be lost or changed. For example, I’ve just been working on photos of a grass where hairs are important for identification. Sometimes I stare at the photo wondering if those hairs are present but in poor focus or just not there. It wouldn’t take much manipulation to accidentally add or remove a hair or two and thus change the ID I might make.

2 Likes