• Welcome to BirdForum, the internet's largest birding community with thousands of members from all over the world. The forums are dedicated to wild birds, birding, binoculars and equipment and all that goes with it.

    Please register for an account to take part in the discussions in the forum, post your pictures in the gallery and more.
ZEISS DTI thermal imaging cameras. For more discoveries at night, and during the day.

Auto-ID of birds in photos—now for UK and Ireland (1 Viewer)

I didn't guess - I downloaded it and gave it a rough and ready sort of test. I must admit I found it to be pretty impressive, although I did manage to fool it a couple of times. There's a brief summary on my twitter feed. I may get round to running more tests on some more complex groups in the near future, such as the aforementioned acros, or LWH gulls, etc.

https://twitter.com/FatPaulScholes1

While I was impressed with its capabilities, I was left wondering what the future of birding will look like as this technology gets better and better, and becomes more and more commonplace. Will birders like us exist in the future? Why would anyone devote a lifetime to something when the answer is just a click away...
 
Nutcracker said:
No, I was postulating 'what if they were' :t:
Click to expand...

Well, firstly I guess those people making up data were still probably seeing Basra Reed Warblers. Secondly if they were photographing something else then one would hope that peer review in eBird would detect the mistakes. Thirdly images being used in the Merlin training dataset are also checked.... A small number of incorrrect images actually shouldn't throw the algorithms too badly. But sample sizes for taxa like this do matter....

I can see the grounds for technophobia but this is an 'aid'; I haven't thrown my OS maps away even though I have a cell phone with Google Earth. Consider two scenarios.

1) Scenario A. ID guru 'a' is on a beach in Cornwall (with not enough phone reception to get data) on his/her own. (S)he spots an odd gull that looks good for a WP first California Gull. Birds at this site don't stick and its half an hour's walk to the nearest decent reception spot at the top of the cliff. CG is a big call and a second opinion is warranted. A BOC shot into Merlin and it confirms the suspicion and is enough to warrant sticking the proverbial neck out for trial by Birdforum. News goes out, 4 people see it before it flies west never to be seen again. In scenario b, the observer walks to get signal and whatsap the BOC to someone else, no-one else ever sees the bird and the inevitable dissection of metadata occurs.

2) Scenario A. Big Garden Birdwatch participant B sees an odd bird in their garden, its not in his/her RSPB Book of Garden Birds so they take a picture and run it through Merlin. The ID help says its most likely an American Tree Sparrow. (S)he contacts the RSPB sending photos and news goes out the same day. Scenario B, (S)he decides it was probably just a funny sparrow as nothing looks similar in the book and goes back to watching daytime TV.

Also remember that Birdguides/RBA etc exist to 'proof' news - no reason why people having access to help should result in more duff gen, it should however result in more rarities being identified. And more importantly, more people getting into birding.

Z
 
Last edited:
Fat Paul Scholes said:
There's a brief summary on my twitter feed. I may get round to running more tests on some more complex groups in the near future, such as the aforementioned acros, or LWH gulls, etc.
Click to expand...

It only does common UK Acros for the time being, LWH gulls of UK + North America a good challenge for it. It doesn't know Slaty-backed yet though....
 
I'm definitely going to give it a whirl - sounds fun and it's not like it costs an arm and a leg (says the person with way too many field guides).
 
Fat Paul Scholes said:
While I was impressed with its capabilities, I was left wondering what the future of birding will look like as this technology gets better and better, and becomes more and more commonplace. Will birders like us exist in the future? Why would anyone devote a lifetime to something when the answer is just a click away...
Click to expand...

Precisely my worry.

As already said, I do use mobile phone and satnav despite recognising they aren't perfect... I will probably use this (I might find it useful abroad rather than here).

John
 
Fat Paul Scholes said:
Why would anyone devote a lifetime to something when the answer is just a click away...
Click to expand...

But what does that mean? You expect people to give up because its too easy? Historically we didn't have access to a reference collection of nearly 4 million bird photos on one site https://ebird.org/media/catalog or access to satellite imagery from the entire planet https://www.google.co.uk/intl/en_uk/earth/ or access to all the specimens, natural history data or real time weather data, bird news information and all the other things available at the click of a button.

This has made birding better in my humble opinion....
 
Surreybirder said:
I wouldn't presume to question your shrike ID's, Paul, but you have some lovely photos there.
Click to expand...

You are too kind.

ZanderII said:
So you are saying those aren't Basra RW? Are these the authors of the fabricated data papers?

This seems like pretty lazy criticism all round.... why not have a go at testing Merlin... and re-identifying the Acros if need be.
Click to expand...

In any event, I recall though haven't checked that the issue was fabrication of breeding behaviour rather than misidentification?

Fat Paul Scholes said:
I didn't guess - I downloaded it and gave it a rough and ready sort of test. I must admit I found it to be pretty impressive, although I did manage to fool it a couple of times. There's a brief summary on my twitter feed. I may get round to running more tests on some more complex groups in the near future, such as the aforementioned acros, or LWH gulls, etc.

https://twitter.com/FatPaulScholes1
Click to expand...

Really interesting - thank you.

All the best
 
Regardless of the huge leaps in the information available to us, critically, up until now, the final decision has always been up to the observer. This changes that - and with it, potentially it takes away opportunities to learn, reasons to engage with other birders and the birding community in general. That's got to be a bad thing?

It won't change the way that you or I bird - but I grew up enjoying learning, developing, and still do both. I just think that something that ultimately will reduce thought, discussion and engagement is going to be less conducive to the development of future generations of useful birders.
 
Fat Paul Scholes said:
Regardless of the huge leaps in the information available to us, critically, up until now, the final decision has always been up to the observer. This changes that - and with it, potentially it takes away opportunities to learn, reasons to engage with other birders and the birding community in general. That's got to be a bad thing?

It won't change the way that you or I bird - but I grew up enjoying learning, developing, and still do both. I just think that something that ultimately will reduce thought, discussion and engagement is going to be less conducive to the development of future generations of useful birders.
Click to expand...

Mark

It could also become an opportunity to learn more quickly if it gave you tips, pointed out the identification features, fed back on your identifications, etc. It is all a case of how it develops.

All the best
 
Fat Paul Scholes said:
Regardless of the huge leaps in the information available to us, critically, up until now, the final decision has always been up to the observer. This changes that - and with it, potentially it takes away opportunities to learn, reasons to engage with other birders and the birding community in general. That's got to be a bad thing?
Click to expand...

The decision rests with the observer as to whether to believe the information provided. You have thumb through the app like a fieldguide when faced with choices... Pretty sure I'm still going to be interacting with other birders whilst having a machine learning field guide in my pocket.

Fat Paul Scholes said:
It won't change the way that you or I bird - but I grew up enjoying learning, developing, and still do both. I just think that something that ultimately will reduce thought, discussion and engagement is going to be less conducive to the development of future generations of useful birders.
Click to expand...

I still can't see why, any more than field guides might have stopped 'discussion', in stopping the monopoly of ornithologists and experts pre their invention. Its another tool in the arsenal. Moreover if you went to Snettisham you aren't going to photograph all the birds present. You still have to recognise something as unusual in the masses. This comes with learning the common birds first. That process may involve Merlin but it has to be iterative. Otherwise you would never get better...

Looks like you caught it out a few times anyway...
 
ZanderII said:
The decision rests with the observer as to whether to believe the information provided. You have thumb through the app like a fieldguide when faced with choices... Pretty sure I'm still going to be interacting with other birders whilst having a machine learning field guide in my pocket.

I still can't see why, any more than field guides might have stopped 'discussion', in stopping the monopoly of ornithologists and experts pre their invention. Its another tool in the arsenal. Moreover if you went to Snettisham you aren't going to photograph all the birds present. You still have to recognise something as unusual in the masses. This comes with learning the common birds first. That process may involve Merlin but it has to be iterative. Otherwise you would never get better...

Looks like you caught it out a few times anyway...
Click to expand...

A field guide doesn't tell you what you are looking at. It makes you look at details, interpret them in conjunction with the guide and learn simultaneously the features and how to interpret birds in different conditions with field guide data.

This device just gives you an answer. No learning. No details. No remembering. Dependence on the device, just like all the people that no longer know their friends' phone numbers, or drive into rivers/railway lines due to duff satnavs.

This thing will substitute for learning for most people. Ability will decrease.

Edit: Here's a thought. We've already seen the committees reject a putative first for Britain due to no photo. How long before they start asking what Merlin makes of the photo? Alternatively, with the availability of Merlin, what need for committees?

John
 
Last edited:
Farnboro John said:
A field guide doesn't tell you what you are looking at. It makes you look at details, interpret them in conjunction with the guide and learn simultaneously the features and how to interpret birds in different conditions with field guide data.
Click to expand...

1st off, have you actually downloaded the app?

One part is a step-by-step key that walks you through your description of the bird. The other is machine learning photo ID. I presume the main problem is the latter. When you stick your image in it throws up suggestions. Multiple suggestions ranked by what it thinks is the most parsimonious ID. With these is a text and more images of the bird in question - exactly like a fieldguide describing features etc. If you are convinced by the ID you tell the machine that this is your bird. I.e. you identify it, that way the program and you get to learn stuff.

Farnboro John said:
This device just gives you an answer. No learning. No details. No remembering. Dependence on the device, just like all the people that no longer know their friends' phone numbers, or drive into rivers/railway lines due to duff satnavs.
Click to expand...

See above - there are 'details'! How many drivers do you know that drove into a river? Is this a common thing or are we going all Daily Mail?


Farnboro John said:
This thing will substitute for learning for most people. Ability will decrease.
Click to expand...

One of the cool things about science is we can make hypotheses and test them. I don't believe that people will get worse, you do. And that can be tested soon, cf http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0139600

Farnboro John said:
Edit: Here's a thought. We've already seen the committees reject a putative first for Britain due to no photo. How long before they start asking what Merlin makes of the photo? Alternatively, with the availability of Merlin, what need for committees?
Click to expand...

I'm not opening up the 'need an image' can of worms as that one has been done to death. The BOURC won't be affected so we are talking local and national (BBRC) rarities committees. Committees still need to decide provenance, decide whether fraud is involved and whether or not stuff could be hybrids (those also needs to be included in Merlin). Some pics may be too cr@p even for Merlin.

Why not ask Merlin? Especially if we could get some quantitative data about the whys of an ID, rather than a black box answer. Identification maybe art and science, why not make it more science. DNA-based ID is now a part of our lives, some don't like it, fair dos. Machine-learning is here to stay, iNaturalist weren't far behind Merlin and the eBird team in launching their software.

Exciting times.
 
ZanderII said:
1st off, have you actually downloaded the app?

One part is a step-by-step key that walks you through your description of the bird. The other is machine learning photo ID. I presume the main problem is the latter. When you stick your image in it throws up suggestions. Multiple suggestions ranked by what it thinks is the most parsimonious ID. With these is a text and more images of the bird in question - exactly like a fieldguide describing features etc. If you are convinced by the ID you tell the machine that this is your bird. I.e. you identify it, that way the program and you get to learn stuff.



See above - there are 'details'! How many drivers do you know that drove into a river? Is this a common thing or are we going all Daily Mail?




One of the cool things about science is we can make hypotheses and test them. I don't believe that people will get worse, you do. And that can be tested soon, cf http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0139600



I'm not opening up the 'need an image' can of worms as that one has been done to death. The BOURC won't be affected so we are talking local and national (BBRC) rarities committees. Committees still need to decide provenance, decide whether fraud is involved and whether or not stuff could be hybrids (those also needs to be included in Merlin). Some pics may be too cr@p even for Merlin.

Why not ask Merlin? Especially if we could get some quantitative data about the whys of an ID, rather than a black box answer. Identification maybe art and science, why not make it more science. DNA-based ID is now a part of our lives, some don't like it, fair dos. Machine-learning is here to stay, iNaturalist weren't far behind Merlin and the eBird team in launching their software.

Exciting times.
Click to expand...

If you say interesting I will agree with you. I haven't downloaded the app, and shan't because others who have, have already said on here that it is not ready for release (i.e. makes mistakes as soon as something a bit difficult pops up.) It needs more work. I get given too much unfinished software at work, I don't need it at home.

But I can only be walked through a description of a bird because of my extant level of learning and observation. And that's where this tool will fall down and lead to bad science: the creation of false positives by incompetents, swamping the capable. Show me a machine that is capable of saying "I can't make bricks without straw" (i.e. give me something reasonable to work on!) and we might be getting somewhere.

I like DNA based ID because it is credible (though I reserve the right to doubt workers who skimp their work, often by looking only at mitochondrial DNA rather than nuclear). I will like this tool when it becomes similarly infallible. But the premature release of incomplete works ought to be penalised somehow. Maybe the committees would like to indicate a level of uncertainty/error that they would find acceptable in the scientific record. Ideally app developers would then have to demonstrate compliance before issue.

John
 
Farnboro John said:
If you say interesting I will agree with you. I haven't downloaded the app, and shan't because others who have, have already said on here that it is not ready for release (i.e. makes mistakes as soon as something a bit difficult pops up.) It needs more work. I get given too much unfinished software at work, I don't need it at home.
John
Click to expand...

The limitations are in the number of images, show it a baby bird and its struggling. All it needs is more images. I would suggest that its already much better at identifying Nearctic waders than 99.5% of BF contributors. I think that is pretty impressive and deserves more credit than your 'not ready for release attitude'.

Farnboro John said:
But I can only be walked through a description of a bird because of my extant level of learning and observation. And that's where this tool will fall down and lead to bad science: the creation of false positives by incompetents, swamping the capable. Show me a machine that is capable of saying "I can't make bricks without straw" (i.e. give me something reasonable to work on!) and we might be getting somewhere.
John
Click to expand...

Didn't understand this bit. Swamping the capable? Is this really going to happen?

Farnboro John said:
I like DNA based ID because it is credible (though I reserve the right to doubt workers who skimp their work, often by looking only at mitochondrial DNA rather than nuclear). I will like this tool when it becomes similarly infallible. But the premature release of incomplete works ought to be penalised somehow. Maybe the committees would like to indicate a level of uncertainty/error that they would find acceptable in the scientific record. Ideally app developers would then have to demonstrate compliance before issue.
John
Click to expand...

Compliance with what? That is like saying we need to hold field guides to account if the number of primaries in a plate is one too many. Infallible? Is that really going to be the case? Ever? Flights shots of cryptic Phyloscs?

Come back and talk when you've had a go John...
 
Out of curiosity, what does it make of this pic? I know what it is.

Edit: forgot to say, the pic is unedited apart from cropping from the original.
 

Attachments

  • IMG_2103 copy.jpg
    IMG_2103 copy.jpg
    681.7 KB · Views: 161
Last edited:
Surely if the scope is within the limited scope of regular UK species, it would be best to start feeding it photos of the species normally differentiated by comparatively subtle features to see how it handles those? Eg. Marsh/Willow Tit, Chiffchaff/Willow Warbler, Herring/Yellow Legged/Caspian Gulls, Crossbills?
 
Warning! This thread is more than 7 years ago old.
It's likely that no further discussion is required, in which case we recommend starting a new thread. If however you feel your response is required you can still do so.

Similar threads

01101001
Common Magpie subspecies
Replies
8
Views
935
01101001
01101001
H
Review of birding apps I use on Android
Replies
7
Views
2K
THE_FERN
T
TicoTyler
Target List Generation
Replies
0
Views
358
TicoTyler
TicoTyler
01101001
Gallery, Bush Warbler in breeding plumage, China, 25 May 2008
Replies
0
Views
269
01101001
01101001
C
AX Visio: is it really as bad as some reviewers claim??
2 3 4
Replies
68
Views
7K
Canip
C

Users who are viewing this thread

Total: 2 (members: 0, guests: 2)
Back
Top