Magic Lantern... CONTINUOUS raw recording @ 24fps on 5D3

tickerguy wrote in post #15931436
There's no value to that in a still camera.

???

tickerguy wrote in post #15931436
There's no value to that in a still camera.

Well, it is for those who shoot video. Which the sub-forum that you are commenting in is dedicated to.

Sweet. Gonna let some reviews/comparisons roll out for the new firmware on the 5D2.

tickerguy wrote in post #15931436
There's no value to that in a still camera.

You are missing something here.

Canon have deduced that a large percentage of their customers expects video in the camera. Even if the user will never ever really use the function, it's a check mark that is seen as required.

So Canon have added video.

But video is very sensitive to aliasing - more so than stills because you see multiple images with small changes presented directly after each other.

The solution to that is that Canon adds a bit stronger low-pass filters. Which slightly reduces the image quality for stills.

If Canon can record video by reading out every single pixel and do an on-the-fly rescaling then the video guys gets the advantage that they get lower noise because they get larger "pixels". And they get to use the full sensor size.

And since the video pixel in the output stream is formed by multiple pixels from the sensor, you don't get as much aliasing as if you use the native, tiny, pixel size of the sensor.

So a camera that can rescale video from the full sensor can make do with a low-pass filter that is optimized for stills. Hence we will no longer need the compromise with a bit too strong low-pass filter.

That is until they start record in 4k, in which case the video resolution will once more be very close to the stills resolution of the sensor, which means that a 4k video will not be able to do much averaging over pixels so there will once more be a need for a stronger low-pass filter for video than what you want for stills.

I'm not missing it at all.

Run the math on reading the entire sensor on a 5d3 @ 24fps and tell me what the internal data rate that has to be sustained -- and the rate at which data must be processed and delivered to the card -- for the various potential options available in terms of chroma (e.g. 4:4:4, 4:2:2 or (yuck!) 4:2:0) and depth of each pixel (number of bits.)

For 1080p as a center-crop you're WAY beyond the limit of what the camera has to deliver for still photography which is why people figured it was impossible. It turns out it's not. But if you think you can do it with a full-frame sensor and then interpolate down in real-time, well..... run those numbers first and let me know if you still think that's reasonable.

There is a reason that high-def (and 4k) cameras don't do it that way and it has to do with the fact that trying to process 5760 x 3840 x 3 (RGB) x 14 (bit depth) / 8 (bits/byte) * 24 (fps) = holy crap.

tickerguy wrote in post #15932653
I'm not missing it at all.

Run the math on reading the entire sensor on a 5d3 @ 24fps and tell me what the internal data rate that has to be sustained -- and the rate at which data must be processed and delivered to the card -- for the various potential options available in terms of chroma (e.g. 4:4:4, 4:2:2 or (yuck!) 4:2:0) and depth of each pixel (number of bits.)

For 1080p as a center-crop you're WAY beyond the limit of what the camera has to deliver for still photography which is why people figured it was impossible. It turns out it's not. But if you think you can do it with a full-frame sensor and then interpolate down in real-time, well..... run those numbers first and let me know if you still think that's reasonable.

There is a reason that high-def (and 4k) cameras don't do it that way and it has to do with the fact that trying to process 5760 x 3840 x 3 (RGB) x 14 (bit depth) / 8 (bits/byte) * 24 (fps) = holy crap.

Who says the info has to be processed? As I understand it they are just dumping the raw data into the cards. If Live View can process the image from the whole sensor into something I can watch on my tablet or laptop at 24fps I don't see why just dumping the data straight to a card somehow requires more processing?

tickerguy wrote in post #15932653
I'm not missing it at all.

Run the math on reading the entire sensor on a 5d3 @ 24fps and tell me what the internal data rate that has to be sustained -- and the rate at which data must be processed and delivered to the card -- for the various potential options available in terms of chroma (e.g. 4:4:4, 4:2:2 or (yuck!) 4:2:0) and depth of each pixel (number of bits.)

For 1080p as a center-crop you're WAY beyond the limit of what the camera has to deliver for still photography which is why people figured it was impossible. It turns out it's not. But if you think you can do it with a full-frame sensor and then interpolate down in real-time, well..... run those numbers first and let me know if you still think that's reasonable.

There is a reason that high-def (and 4k) cameras don't do it that way and it has to do with the fact that trying to process 5760 x 3840 x 3 (RGB) x 14 (bit depth) / 8 (bits/byte) * 24 (fps) = holy crap.

So then you are missing this sentence in the post you commented on: "Now we just have to wait until Canon releases new hardware that is fast enough to do a full-sensor read at full video speed [...]".

Notice the "releases new hardware" part? I'm not talking about 5D2 or 5D3. So it doesn't matter what bandwidth calculations you do for a 5D3.

In the end, bandwidth doesn't matter much, when you design sensor hardware. Why? Because bandwidth gets divided by number of read channels. Meaning it is possible to have a large number of concurrent transfers, demosaicing, rescaling ongoing. So it is only the final step - bandwidth to memory card - that remains. And there, it is still possible to perform non-lossy data compression.

By the way - your example for full sensor data bandwidth was based on RGB. But the sensor isn't RGB. So raw capture don't need that x3 factor.

Thought ML said it will most likely come to the 6D as well. Possibly even the 600D.
Aren't SDXC comparably fast? There are SD cards that read 100MB/s

Kentaro wrote in post #15932800
Thought ML said it will most likely come to the 6D as well. Possibly even the 600D.
Aren't SDXC comparably fast? There are SD cards that read 100MB/s

It's not about read speeds it's about write speeds. CF kills SD in write speed.

Scatterbrained wrote in post #15932871
It's not about read speeds it's about write speeds. CF kills SD in write speed.

ah right, oops. Write speeds are still up to 70, 80MB IIRC. I would think that is sufficient. Shooting 3, 4k seems a bit unreliable at the moment anyway. Would like to see more real world application to 2k raw pre/post process.

pwm2 wrote in post #15930664
When you have stopped laughing, you might consider the difference between the skills of the Canon developers (who just might be the same developers who implements the video functionality on their professional video cameras) and what product management decides that they may implement.

Realize that it isn't the engineers skills but product management who decides how to split the market, i.e. what features that should separate the different camera models. Canon don't want their DSLR to get too close to their professional video cameras since they are in a different niche with a totally different price.

That is the main reason why we need stiff competition, so the product managers will feel a pressure to keep up, instead of just deciding on extremely high prices in the belief that a Canon is unique and not suffering from competition.

I understand, but dont care about any of the above.
I dont understand why people are willful Canon apologists and rationalize their stone age business model which delivers crappier products than they could or should.

If you're going to argue they have to make things crap in order to preserve a complex heirachy of cameras, then dont. Axe some of the line up.

If you make all of your products ****ty for sake of competing with yourself then argue that software costs too much to develop innovative features for any of the individual cameras, that doesnt make any sense at all.

As ML has demonstrated, there's a massive amount of software side functionality waiting to be untapped in even an entry level DSLR.
Imagine if a Cell Phone company stagnated in the same way, they would be out of business in 3 seconds flat.

ML is certainly the only reason I bought a Canon.

romanv wrote in post #15933279
I understand, but dont care about any of the above.
I dont understand why people are willful Canon apologists and rationalize their stone age business model which delivers crappier products than they could or should.

Then you still have lot to learn about business models.

First off - your previous post did contain a "lol" before you told how lousy you thought Canon developers was. Which clearly shows that you did not understand the concept where the product management decides that features to ship.

From your new post, it sounds like you base your view on the idea that Canon exists to make the customers happy. But the main reason for Canon to exist is to produce a profit to deliver to the owners.

And one important route there is to create product stairs, where people will have to pay extra to get a one step bigger/better product. And where a customer who have already bought a product can be tempted to come back as a repeat customer to upgrade to a better product - hence having been tricked into paying for two products. This is extra valuable since lots of customers will keep the older/smaller product even after having upgraded, avoiding having the older product compete on the market.

It isn't a stone age business model, but a model that is well proven in battle. It is a business model that works, and works well. Which is the reason why it is in effect in so many different areas and by so many companies.

But it is a business model that is normally complemented with a second business model, where you regularly upgrade the line of products. So a new low-end product will slowly overtake an older mid-end product.

This means that a customer gets tricked to either buy a more expensive, high-end, product. Or that the customer gets tricked into buying a newer product of the same tier.

But all the time, the customer gets tempted to pay more money. Either to pay for more than they actually need, or for replacing what they already have.

Calling it stone age is just a big failure in understanding the forces that drives companies - what is motivating them. They are not motivated by you feeling happy about the product you just bought. They are motivated by you digging deeper in your wallet, or more often opening your wallet. And their method of making you pay do work well - no one have actually managed to find any better business model. Maybe you are the new Nobel laureate, being able to show a more profitable business model - but then you have to start by actually describing, believably, what the companies should do to increase their profit more using an alternative model.

Making things crap? That is a huge failure in your analysis. Compare todays cameras with the cameras you could buy 5 or 10 years ago, and you'll quickly see how badly your analysis failed.

Todays cars are safer if you have an accident, and they draw less fuel.
Todays cameras have better dynamic range and resolution while offering live view, video etc.
Todays audio systems gives you better sound quality and more audio channels despite being sold at a lower price.

Using the word "crap" if a company doesn't ship the absolutely best they can manage just shows that you don't understand the word "crap". And that you also fail to understand product development costs, versus expected sales price.

A company that only sells the best they can produce would be forced to charge a much higher price. On one hand, because it would be a high-end product so the majority of people wouldn't be able to afford it. So that high-end device would sell in fewer numbers requiring the price to be set very high to cover all R&D costs. So that would mean even fewer people would be able to afford it. Which means the price needs to be even higher. In the end, a company settling for only selling "the best", would have a very small customer base. The type of customers where "price doesn't matter".

But more importantly - such a company would not have the "sales pump" that would motivate them to constantly work with new releases. They might instead sell their "high end" product for many years while they work on the next magic wonder. In the end, the technical progress in society would be much slower - wouldn't you say that would be crappy?

In the end, what drives technological progression is the need to constantly release new, slightly improved, models to drive people to update. And the need to offer products that are good enough compared to the competitors offerings.

For that second part, it matters much if you have a fan boy customer base or not. When you get religious customers, you can sell your products even if your products aren't really good enough to compete. And if you have products where the customer gets invested and can't easily switch, then you can also manage to be a bit slow to respond to competition.

But in the end, it will be competition and the customers need to upgrade that is the main driving force for how much new functionality you will see in newer cameras. This means that the engineers can often have new - and better - designs available. But it isn't in the company's interest to sell that functionality today. It is better saved for tomorrow, when you release your updated model line. Maybe just to get your customers to update. Or maybe because you need that update then as response to the competition.

Of course this post will not matter much to you. You have already indicated that you don't think the Canon engineers knows much about video, despite Canon being one of the big companies in professional video.

So you aren't likely to get in a position where you will be product owner and decide on model lines in a while. At least not until it starts to make sense, that you make more money by having multiple model lines - which implies that you need to avoid giving some functionality to some of the products to differentiate them. Without differentiation - what would then motivate the difference in price?

Not really - you are failing to notice that there are different product segments with different driving forces. And your comparison also seem to relate to the top-of-the-line offerings from the phone manufacturers.

Lots of phone manufacturers sells the same phone as multiple different products. They change some menu options and the color/shape of the shell and release as a new product. Not only that - but some few companies sells the same software for use in a large number of different phones. Just with some branding cosmetics. So you, the customer, thinks you bought a different phone. But it was basically the same phone even if bought from a different phone manufacturer. But inside, it's the same chipset as used by several other phone manufacturers. And it's the same software too. Except for that rebranding front end.

The only real innovations in the phone industry happens in their top-of-the-line models, where they produce new chipsets, that they might produce for maybe 18 months and then kills off. Look for example at processors etc from Samsung. You can buy their embedded processors for use in your own equipment - but only if you buy and make some 100k to millions of devices during a short time span. Then you need to redesign your product because you can't get that processor anymore but need to use the next generation processor instead.

The leading-edge mobile phone industry lives in a market segment where they sell a high-end product that the customers uses for a very short time before throwing the product and upgrading to something new. That is ok, since the price of even a high-end phone is quite low. Look at the production cost for full-frame high-end sensors to a DSLR, and you will quickly realize that not too many customers could afford to buy a high-end DSLR and then replace it every 12-24 months. So trying to enforce the same development cycle for DSLR as for mobile phones is a failure.

The manufacturers have zero problems running 12 month product release cycles for their camera models, be it entry level or high-end. Just that the price and the customer base will not be able to buy the cameras using a similar pattern as cheap mobile phones.

If you a new entry-level DSLR is released every 12 months, the customers will not upgrade their camera every 12 months anyway. The customers will upgrade using a staggered pattern, where they would upgrade maybe every 3 releases. This can be acceptable for a camera manufacturer, since we are talking about a high-volume (relatively speaking) product. But for the high-end DSLR, the number of sold units gets too low. And the cost for marketing material etc will get too high. So it would not make sense to try any 12-month cycles for 1-series cameras. Especially since product cycles for 1-series cameras requires real R&D for every new release. While the entry-level cameras can get spin-off functionality.

And one thing here then is that the DiGIC chips contains a lot of functionality. Not all of it used in all sold camera models. But it isn't economical for Canon to try to strip out functionality from the chip for specific camera models. So the chip is just as powerful when used in a low-end or high-end camera. Just that the high-end camera might potentially get two DiGIC chips for extra processing power.

This isn't different from a car manufacturer who might have a big factory for making transmissions. So it can be economical to use the same transmission in multiple car models, even if one car model have a much weaker engine that doesn't need such a robust transmission.

You wouldn't find it strange if someone tunes their car to get 100 bhp extra and the transmission still works well. But you feel Canon is a broken company for doing stupid things when ML plays with the software and makes use of that little extra available inside.

Well, everyone is free to decide their own reasons for buying a product. But in the end, you would find that Nikon is similar to Canon. They also release models that aren't the best they could be. You need to select a camera brand that don't care about the masses and only sells top-of-the-line products if you want to find a difference. So the question then is: Why haven't you selected Leica? Maybe their price and their video support isn't to your liking?

So because Canon wants to make a profit, they dont care what customers think?
I thought their customers were the ones spending money to make them a profit.

Regarding cell phones, I wasnt so much talking about hardware so much as their push to add as much functionality to a given cell phone's hardware as possible.

I own a cheap ****ty Samsung cell phone.

I can connect to the internet with it, and download bazillions of different user made applications, some of which will ring the absolute neck out of the phone's hardware.

None of which discriminates against having a cheaper phone, unless the hardware cant keep up. If I want to run some of the more intensive programs etc, then I buy better hardware.

Even better, the phone manufacturer doesnt have to do anything. Apart from making the phone Android compatible. The users profit by selling apps, the manufacture profits by selling hardware.

My phone can become a GPS unit, navigation system, lap timer, blah blah. Pretty much only limited to the imagination of some users.

If Canon were to make a cell phone right now, it could probably call people, but then you'd have to buy the $200 text pad if you want to send text messages. Oh, you want to download an app? Sorry mate, this is a phone not a laptop geez.

I've no doubt that DSLRs could benefit from Android type of functionality - as ML has demonstrated... But without any wilful assistance from the manufacturer.

Canon decides that I shouldnt have an intervalometer, the most basic software routine in the universe, on a $1000+ camera.
When even a cell phone can do that, if you download an app. DSLRs arent sold in a vacuum, they are competing against other electronic goods, and also what people expect from electronics is changing on account of Android / Iphone type devices / tablets etc.

My cell phone can stitch photos together automatically - my DSLR cant.
My Cell phone can download an app to have an intervalometer - my DSLR cant.
And so on and so on.
Some people are bound to ask, if it's a top end product then why does it lack features which even a non camera (cell phone) is capable of?

The point and shoot market has suffered terribly on account of this.

Non of the above which you've said addresses why Canon or any other manufacturer, cant incorporate even a simple Intervalometer function into a digital camera.
This function isnt even available on the top of the line models, as best I know, so it's not as though they're protecting that portion of the market.

Same goes for zebras, why doesnt even their top of the line camera have basic zebra functionality for blown highlights etc? Easiest thing in the universe to incorporate into a camera, yet none do. So the argument about protecting upper segments doesnt make sense. Neither does operating as though it's only Canons other products that they need to compete against.

As you say, if a line of cameras share the same Digic processor or whatever, there's no reason to add a compatible software feature to all products on a line.
Having something like an intervalometer could be the tipping point between someone choosing Nikon or Canon, for the expense of getting someone to spend 1/2 a day writing a software routine.

When I was first looking to buy a DSLR, I wrote a list of all the things I wanted it to do, then went looking for what options fitted that. Which was pretty much none, because none could do a time lapse without spending $$$ on a box that plugs into the side. Which seemed rather baffling.
So no one got my money.
Until I found out about ML, then bought a 600D.

Either way, Canon's internal politics has absolutely zero to do with me enjoying taking pictures, and I have zero interest in learning about what fuels their laughable innovations compared to what a few guys hacking around in the firmware can acheive.

I just want a camera that has a bunch of different useful features, and a 600D + ML has kept me more than happy. I'm more interested in their ongoing developments, than anything Canon's done in the last while.

romanv wrote in post #15933601
When I was first looking to buy a DSLR, I wrote a list of all the things I wanted it to do, then went looking for what options fitted that. Which was pretty much none, because none could do a time lapse without spending $$$ on a box that plugs into the side.
So no one got my money.
Until I found out about ML, then bought a 600D.

Haha good luck finding a camera that does everything you want it to do.
However if you can justify spending $1000+ on a body, I'm sure you can crunch out another 30 bucks for a box.

What was that about RAW recording?