Re: Electronic or mechanical shutter?
Distinctly Average wrote:
Ephemeris wrote:
Distinctly Average wrote:
Ephemeris wrote:
MarshallG wrote:
Ephemeris wrote:
Distinctly Average wrote:
Ephemeris wrote:
Distinctly Average wrote:
Ephemeris wrote:
Distinctly Average wrote:
Ephemeris wrote:
I've seen some software that automatically tires to correct for e shutter creates problems in videos.
Do we think some (maybe not all) could be repaired automatically either with an external tool or in camera?
Personally I doubt rolling shutter could be corrected that way. Capture is at a different moment, so any correction via AI would really be a false representation of the scene.
-
Given it is done for Video would it not be plausible?
False/not false - isn't that a little bit of a moving definition?
-
Just enjoying photography
https://www.distinctlyaverage.co.uk/
https://www.flickr.com/photos/distinctly_average/
How would you re-construct missing dragonfly wing parts for instance? In a stills shot you want all the detail but you n video you can get away with parts that are less than accurate. It is also probably easier in video as there are multiple frames to get data from.
I think before deciding if something is impossible we could have a try at making the problem less so.
Let's say I take an image of a vehicle and it has crooked light posts in the background then I can correct it. So it must be possible.
As for missing parts well AI tools do some pretty darn good things with that also. We can see that even with Google Pixel phones.
Less than accurate? Isn't that the majority of our images? If you happen to have to photograph geometric like objects frequently they often need a bit of nudging back into shape.
I know we don't have the ability to do it and it's not something that has ever been asked of us but surely those Nvidia, Google like boffins could give it a whirl?
Some snazzy software to improve rolling shutter so that we don't need extremely expensive global shutters or the like?
Not suggesting it is impossible, just very difficult compared to video. Obviously that depends on the image. The AI model would need a lot of training for each and every subject and getting the image sets required is going to be a legal minefield that is yet to play out, It will almost certainly be possible one day, but by then will we actually have to deal with rolling shutter anyway? I personally don’t think that in 5 years it will be something we will be worrying about.
We started with doubt it could be done, to a 5 year requirement.
Can the rolling shutter issue be negated now? Maybe but I've not seen it at high frame rates (so can it be done at say 240fps).
Cost is a factor here and market places. The market places include mobile phones, security cameras and automotive vehicles as well.
Could we make improvements to the system now? Given it sounds plausible I'd think so.
Can I do anything about it? Very unlikely to be a big contributor Vs the might of the Googles.
I'm glad we do have the Google's and the Musk's becuase can't doesn't seem to be at the forefront.
Can we build a low cost camera, with reasonable resolution, frame rate that has multiple exposures across a frame? Absolutely and it's already in a luxury vehicle brand. I remember being told it couldn't be done - Panasonic told us that.
Anyhow, I'm absolutely not disagreeing with you as we are both just thinking in the wind. I also hope in 5 years this problem isn't one (however it's acheived).
Nikon and Canon’s most expensive cameras have sensor readout that’s so fast that there is no rolling shutter. Nikon’s is so fast that they don’t even have a mechanical shutter.
There are unwanted side effects to Nikon's solution and Canon is a tad pixel light. So more side effects.
So I agree it’s a very safe bet that the R5 and R6-level cameras will not have rolling shutter issues in 5 years or less. Of course, someone will inevitably find a corner case, like the shape of bokeh balls when the lens is wide open and the shutter speed is super high, something like that.
However the market is a lot bigger than the R1/3/5 world and maybe some of the side effects are really quite tough to fix.
The R3 still doesn't manage to take images of rotating fan blades without problems so even it could benefit of my pipedream 🤗
That is probably the biggest hurdle, cost. The companies that could do it would need quite an investment. The return on that investment would be doubtful especially given rapid advances in technology that will inevitably remove the problem. At the moment the tech for fast readout already exists, but is probably a little too expensive for the lower models. But we all know how quickly this kind of technology comes down in price as volumes increase.
I am sure big investors could see this and therefore a specific AI would not be on their radar.
Your logic is a little odd. Given this exists already, commercially for video and yourself talk of 5 yearsj, too expensive etc.
Could it be possible this already exists within mobile phones at some level?
Why would improving an image due to an effect that is an inheritant part of the pipeline not be something being developed?
How would you then speak of the Topaz system which can make improvements to images due to camera shake or out of focus? And does it well (we use it).
Would you have said the same about AI noise reduction?
It it was a side effect of image recognition or the line then we have a maybe
All very different things. What we are talking here is temporal correction. Think of that reflection image from Duane Paton with the bird in a very different position in the reflection than the bird in the air. Noise reduction and minor distortion is not too hard. But when you looke at how AI performs with making people smile etc and it looks very weird in a still. Less so in video as the brain corrects for certain things. It can still be done very poorly, look as superman’s mouth in that justice league film for a prime example, the worst moustache removal in the history of film.
I don't think I can agree in sorry.
If we have a 30fps, or even faster 40 it doesn't matter so much becuase we are using ES. What we have is a series of frames taken in time. Each frame itself also has a time component which is what we are saying is rolling shutter.
Sounds a great place for a computer to fix or at least improve the image.
So it is done in some ways already by Topaz on stills.
It is done by multiple vendors on video.
I manually correct if needed.
That suggests we have growing evidence for it being plausible.
Further more we have images created without a camera in the first place (well I suppose it does need images in the pipeline at some point in the process).
Why do I say expensive? From the point of view of an investor, what would my long term return be and would it grow? I think it would be a hard sell.
Same as the growing range of image improvement software vendors. Let say Adobe, DxO, Topaz etc. They seem to be able to make a living from image enhancement and find it relatively easy to have a large user base (two of them take our money yearly).
