AX Visio: is it really as bad as some reviewers claim?? (1 Viewer)

[Canip]

Anything less than 400mm equiv would be mostly useless to me, unless using a very high MP sensor where crop factor could provide 600-800mm effective.

After some more days getting used to the Visio , I think the current focal length satisfies up to a limited distance, but something like 400mm equiv would be welcome as it would allow to do IDs at greater distances than is the case today. Now, even well visible birds are often too far away and therefore too small in the camera for identification in the Visio (you often have more luck with later ID in the Merlin app), even if you see them well and clearly in the binocular.

At relatively short distances (up to around 50 yards), and for large birds in flight up to perhaps about 100 yards, the identification works quite well, esp. in good light. Would be great to be able to double these distances.

Attached are a few more photos with which the Visio did correct in-device IDs, plus one where the distance was too great:

• BT Blue Tit at short distance
• Grey Heron at 60 yards
• Grey Heron 2 at 80 yards
• MB Common Buzzard at 260 yards (not identified).

 

[tenex]

Now, even well visible birds are often too far away and therefore too small in the camera for identification in the Visio (you often have more luck with later ID in the Merlin app)

As you pointed out yourself, if Merlin often IDs the same photo later, the real problem is not distance or focal length.
Thanks for all your reports!

 

[RobKempers]

Here is the difference between the AX Visio and a Canon R with RF100-500mm lens.
Left with 500mm and right AX Visio, so not the image you see through the 10x32 (500mm) viewer but the camera with 260mm.
My other experience with the AX Visio is the distance at which detection takes place, you have to come from a good background if you want to be able to identify a small bird after 30 meters. In short, a beautiful innovation, but don't have too high expectations of the photo quality. It is possible that work can be done on a larger sensor and image stabilization in the longer term.

 

[Canip]

the distance at which detection takes place, you have to come from a good background if you want to be able to identify a small bird after 30 meters.

This is in line with my experience (see my earlier posts about „realistic expectations“).

Swarovski indicates in their instructions that for successful identifications, the bird has to more or less fill the red circle (two different choosable sizes) visible in the display. Small birds at more than 30m fill less than half of the (smaller) circle, so identification is random. Larger birds are more easy to identify at a greater distance (my Visio correctly identifies white storks, grey herons, common buzzards and red kites at up to 80-100 m, sometimes more, both sitting on the ground and in flight).

Many of the photos of non-identified birds are still useful for later identification in the smartphone. So far, I have been able to identify 4 out of 5 birds not picked up in the Visio by later use of the Merlin app in the smartphone, where you can blow up the bird image to a larger size.

The Visio needs a bit of experimenting to get best results. If you observe birds that are most likely too far away for in-device identification, you may even skip the ID attempt in the Visio and simply make a photo with the Visio’s photo function and then use the Merlin app. So make use of the double functionality that the Visio offers.

500mm focal length would be welcome to overcome the limitations of the current 260mm. It would allow the use of the Visio at greater distances. I wonder whether the built in 13 MP would not allow to digitally increase the current FL and still have sufficient photo quality for recording or ID (via software update).

 

[tenex]

I wonder whether the built in 13 MP would not allow to digitally increase the current FL and still have sufficient photo quality

Playing such games adds no information and should make no difference to ID, at least not without AI enhancement, so one could wonder whether that's what's employed in the Merlin app on a phone but not in Visio itself.

 

[abirdingguy]

Playing such games adds no information and should make no difference to ID, at least not without AI enhancement, so one could wonder whether that's what's employed in the Merlin app on a phone but not in Visio itself.

Playing that game could be helpful in indicating where the automated identification tool should focus. Removing unnecessary parts of the image (parts of the image not containing a bird) could be helpful in allowing the tool to detect and then identify a bird. What is perceived as an identification failure on the part of the tool could actually be a detection failure when the bird doesn't occupy a lot of pixels.

 

[tenex]

Playing that game could be helpful in indicating where the automated identification tool should focus. Removing unnecessary parts of the image (parts of the image not containing a bird) could be helpful in allowing the tool to detect and then identify a bird. What is perceived as an identification failure on the part of the tool could actually be a detection failure when the bird doesn't occupy a lot of pixels.

Visio distinguishes between ID failure and detection failure.

 

[abirdingguy]

Are the examples that Canip provided a case where Visio indicates detection failure, but in 4 of the 5 cases, Merlin detects and correctly identifies the bird? I've not used the Visio, but I've used other automated identification apps where a simple image crop made the difference between detection or not.

 

[tenex]

Are the examples that Canip provided a case where Visio indicates detection failure, but in 4 of the 5 cases, Merlin detects and correctly identifies the bird?

I thought few of them were actual detection failures, but am having trouble relocating them to check. Perhaps Canip can answer this question himself.

 

[grackle314]

In the well written and long post 16 Feb 2024, Canip expressed a question about how close to artificial intelligence Swarovski should position the AX Visio.

Digging around on internet, i found descriptions of Merlin using deep convolutional neural network. Bayesian computer models, and computer vision recognition contributed to by Caltech computer sciemce Professor Pietro Peroni, a leader in artificial intelligence. Merlin also credits Swarovski for being part of the Merlin team.

How close to artificial intelligence Swarovski Optik should describe the AX Visio is up to them. Right now I'm impressed how deeply Merlin has improved by using artificial intelligence as part of the suggested identification process. They have huge learning databases.

 

[Canip]

I thought few of them were actual detection failures, but am having trouble relocating them to check. Perhaps Canip can answer this question himself.

What‘s the question exactly?

 

[Canip]

More samples:

• great spotted woodpecker, easily and correctly identified twice in a row, distance 42 yards (laser measured)
• white stork, correctly identified at second attempt (first attempt: "no identifiable bird"), distance 80-100 yeards (estimate)
• common kestrel, correctly identified, distance 60 yards (estimate)

As discussed, a better camera with more focal length would be great, but the present one is quite usable within limitations.

 

[Dyrlege]

More samples:

• great spotted woodpecker, easily and correctly identified twice in a row, distance 42 yards (laser measured)
• white stork, correctly identified at second attempt (first attempt: "no identifiable bird"), distance 80-100 yeards (estimate)
• common kestrel, correctly identified, distance 60 yards (estimate)

As discussed, a better camera with more focal length would be great, but the present one is quite usable within limitations.

I can't really bring myself to complain!
Will post from here as well as soon as the birds come out of hiding. I keep forgetting it's still only February.
Per

 

[tenex]

What‘s the question exactly?

The immediate question is what proportion of failures in your use of Visio have been of detection of a bird at all, vs identifying species. Then we can wonder how much higher magnification in the camera would help, and how.

Digging around on internet, i found descriptions of Merlin using deep convolutional neural network. Bayesian computer models, and computer vision recognition contributed to by Caltech computer sciemce Professor Pietro Peroni, a leader in artificial intelligence.

I wonder whether all that can be happening on your phone; perhaps some of the hard work occurs in the cloud instead, and that's what's missing in Visio. Canip, if you disable its network connection, does the Merlin phone app still outperform Visio on the same photos?

 

[Canip]

More samples:

• great spotted woodpecker, easily and correctly identified twice in a row, distance 42 yards (laser measured)
• white stork, correctly identified at second attempt (first attempt: "no identifiable bird"), distance 80-100 yeards (estimate)
• common kestrel, correctly identified, distance 60 yards (estimate)

As discussed, a better camera with more focal length would be great, but the present one is quite usable within limitations.

what proportion of failures in your use of Visio have been of detection of a bird at all, vs identifying species.

My experience: the great majority of detection failures in the Visio are of the „no identifiable bird“ type (reasons mostly: bird too far away, hidden in branches, bird against bright background). Wrong species IDs occur only occasionally, similar to the way it happens in the Merlin app.

 

[Canip]

As indicated in earlier posts, not only does the Visio not feature a screw hole for mounting on a tripod, its body shape makes mounts of the Berlebach type uncomfortable to use.
My solution for stationary birding situations, where carrying a tripod may make sense: the Swarovski binocular tripod adapter with a primitive elastic "extension".

 

[Dyrlege]

Three indoor images:
Lumix G9 250 mm
Same at 400 mm
Visio

Camera and AX on tripods next to each other. Note: the G9 is an M43, so 35mm equivs would be 500 and 800 mm

Edit: the Visio in on a Berlebach bino-mount, worked all right.

 

[Hermann]

Wow. The difference in image quality is quite astounding, even though you didn't take the shots with the Lumix at full resolution. And the Lumix shots are taken at ISO 3200 according to the EXIF data. The Visio shot at ISO 160 (!) according to EXIF data.

Just shows how badly the tiny sensor of the Visio fares against a decently sized sensor.

Hermann

 

[Dyrlege]

Wow. The difference in image quality is quite astounding, even though you didn't take the shots with the Lumix at full resolution. And the Lumix shots are taken at ISO 3200 according to the EXIF data. The Visio shot at ISO 160 (!) according to EXIF data.

Just shows how badly the tiny sensor of the Visio fares against a decently sized sensor.

Hermann

I tried to be fair: the ax is on auto, so I left the G9 to fend for itself. But I'm still impressed with the ax result: it is for documentation, not photo contests😄
The power of the G9 with the 100-400mm is quite lovely!
Edit: the G9 fully equipped plus the AX is till way less than a Canon F4 600 mm...(price has been mentioned as a factor earlier)
Per

 

[Canip]

Just shows how badly the tiny sensor of the Visio fares against a decently sized sensor.

And you should see how badly the Visio camera fares against really good cameras like e.g. my Leica Q2 (which cost about the same as the Visio).

So don‘t count on the Visio if you are an avid wildlife or landscape photographer looking for a good camera.
The Visio is more for people who have been struggling with digiscoping difficulties forever like me and want an easy-to -carry straightforward combination of good binocular optics with a documentation feature that‘s at least acceptable. But, of course a camera with more FL and better sensor would be welcome.

By the way: in the birding community, other features of the Visio beside bird ID seem almost never discussed. I was recently told by an army instructor in artillery that he had been waiting forever for something like the new „share-discoveries“ function with the compass and inclinometer which makes it much easier to teach his class target identification in mountain terrain.
I could imagine hunters and also birders who observe in groups might find these features useful as well.

Canip