CRAW burst on Canon R7, test, tries, results, photos, comments...

I've moved this thread from Tea Rooms to this forum, as it makes more sense.

The number of times I've used RAW burst can be counted on the fingers of one hand. I always forget it's there, and it takes time to set up on R7. RAW burst also generates massive files.

A combination of Mike B and Billysdad piqued my interest by mentioning CRAW files - Compressed RAW files. Mike B pointed them out, and Billysdad pointed out they are available on a Canon R7.

I decided to give CRAW and CRAW burst a go. Anything to reduce file sizes and decrease the time taken to write to a SD file.

Caveats, as always. I shot the following sequence through manky double glazed windows (why does it always rain shortly after I clean them?), with the subject under a large oak tree, with the added bonus of overcast conditions.


Best used on bright sunny days or bright light, probably best in Tv mode, rather than my favoured Program mode.

DPP4 is pants when it comes to extracting images from RAW and CRAW burst files. Firstly, it produces image 1620x1080 pixels in size rather than 6960x4640 pixels of the original RAW image. There might be a setting I've missed to stop this, but I haven't invested the time to RTFM - I'm allergic to reading manuals. DPP4 doesn't appear to page or swap. All images are held in memory, resulting in 'out of memory' error message. It also screws up Windows 11, as I don't think it releases memory correctly when I close it down. Plus it's slow, and the extraction tool has a rubbish UI.

I found it far easier and quicker to extract images in camera. This process also leaves the image at its original size.

The resulting CRAW burst file seemed much smaller than the RAW burst file, about 50-60% smaller. Writing to SD card was definitely faster - as it's a smaller file. I think the individual CRAW files were smaller. Difficult 

Anyway, enough faff, on with the results. These are uncropped images. You'll also see No Mow May in full bloom.

That's it. Over and done with so quickly. Nine frames out of 48 images saved in burst mode.

One tiny issue with this sort of shot. The bird moves so fast, I can't track it. This means the focus point remains pretty much on where the bird started off. One reason the grass, buttercups and daisies are so crisp and clear. OK, another reason for the slightly blurred Jay is that it is moving fast, and I was using Program mode in less than ideal light. Tv mode with bright sun might have resulted in better sharpness.

Personally, I can't see much signs of rolling shutter. I also photographed a Magpie taking off and flying. Although there wasn't apparent signs of rolling shutter, the test was inconclusive as the bird was a fair distance away. However, Jays and Magpies are not noted for their fast wing beats. You would, though, expect fast movement when taking off, therefore more chance of rolling shutter. The last time I tried RAW burst (a year ago) I did get very bad rolling shutter with a Blue Tit.

Mike B did hint that he had read that rolling shutter had largely been removed by firmware upgrades. I hoped it would, but was unsure. Especially as certain learned gentlemen on the internet said it was impossible due to it being caused by hardware. Well, I shall have to try RAW burst on subjects like Tits or small ducks, whose wing beats are much faster.

90% luck, 5% field craft, 5% camera skills.

  • Hi Angus

    I shoot Nikon so have no experience of Canon, but does the R7 have subject tracking? That should keep the bird in the autofocus box wherever it moves across the viewfinder,

  • Wotcher, PB.

    My Canon R7 has both subject tracking (currently set to recognise animals) and eye tracking. I have both turned on all the time, except for those rare occasions when the AI software gets confused.

    I just did an experiment pointing my camera at some Blue tits and then a Great spotted woodpecker whilst in RAW burst mode.

    Firstly, the R7 does not do eye tracking in RAW burst mode.

    Secondly, although it kind of recognised birds as an object, I don't think it recognised birds as a bird.

    Confused? Essentially, I place the small focusing square (centre point) onto a bird. The R7 detects it as an object and tries to track it. However, it is very easy for the R7 to lose track of the bird and focus on something else, like a flower or pole or leaf. Thus, it has reognised an object upon which I first placed the centre point focus square onto, but not as a bird which it has to track.

    There could be a number of reasons why the R7 does not do eye tracking and proper animal tracking while in RAW burst mode. They are:

    1) I've missed a critical setting

    2) The R7 doesn't have sufficient processing power

    3) The R7 can't work out when a bird is a bird. We've already covered this, but more will follow.

    Expanding on the bullet points.

    1) I don't think I have missed a critical setting, though I might well have. I don't have much choice on the RAW burst mode menu page. This is where Billysdad says you missed summat you eejit.

    2) The R7 is the penultimate base model of Canon's new R series; after they mucked about with some strange mirrorless models initially. I could be wrong, but that's what I saw and why I waited until Canon got itself sorted out and produced the R7.

    As such it probably has less a less powerful processor or processors and less RAM (memory) than their mid and high end models. This means any AI tracking software will either not keep up with any subject changes or its capability has to be reduced i.e. not perform eye detection or subject recognition and tracking. It is possible that models such as the R5, R3 etc do allow subject and eye tracking in RAW burst mode, but I wont buy one of those until I win the national lottery or mega premium bond prize. I suspect this has a much chance of happening as Pink porcines being added to the manifest of air traffic control.

    This leads me on to point 3.

    3) I have to guess how subject recognition works or at least how I would do it.

    • Firstly you start with edge detection. The software has to figure out edges, boundaries, etc. AI is not required for this. Don't ask me how it works or I'll whimper. Needless to say, some very bright scientists and mathematicians worked out the maths for this.
    • Secondly, the software takes the edges it has found and works out shapes i.e. join dots together to give a recognisable enclosed area. Don't ask me how this works or I'll cry. Very clever people worked out how to do this.
    • Thirdly comes the recognising what a shape is. You don't have to use AI to do this, but it is easier if you use AI technology. I reckon a Machine Learning algorithm was used to achieve this. The idea is to 'teach' the AI (ML) software so it learns how to 'recognise', say, birds. Generally this is done by throwing a whole load of test images of birds at the AI system, let it work out if the image is a bird or not, and correct the software if it has guessed incorrectly. When the AI software has achieved a certain pass rate, you then throw a whole load of other images at the AI system which it should then have sufficient 'knowledge' to work out what is and isn't a bird.

    I don't know how a ML system works under the hood as I've not bothered to track down any of the algorithms that do the learning. My daughter, who has finished her final year physics degree project, which is all about ML, might have a better idea of what is under the hood. I did read her dissertation, and understood about 20% of it. Mainly words like 'and', 'the', 'written by', 'internal star formation physics', etc, etc, etc.

    The big problem for the AI software, whether on an R7 or higher model with subject recognition available, is recognising a bird.

    Thus, in my post, photos 1 to 4 the Jay is fairly easy to spot as a bird. Photograph 5 is where the problems start. Even a human being, who isn't a birder or hadn't seen the preceeding photos, would have difficulty in recognising the shape as a bird without a lot of squinting and thought. I very much doubt such an image or something even close to it would be in the training pack used to train the ML system.

    At this point, the R7 or indeed other models, would, I suggest, loose track of the bird, because it can't recognise the shape as a bird. The AI software possibly has a better chance of picking out the Jay from photo 6. However, the bird is moving so fast, with images being captured simultaneously along with all the other complicated processor hungry tasks (like focusing and metering), I reckon the AI software wouldn't have time to detect and lock onto the Jay by photo 6 and beyond; hence it stays locked onto the last thing it could remember i.e. buttercups and daisies. I think a lot more heavy duty processing power is required than is packed into an R7, plus a bigger training pack, for the R7 to successfully keep detecting and tracking a really, really fast move subject, like the Jay launching itself into flight. This is an astonishingly fast reaction, and has developed to escape predators. Pigeons have got an amazingly short take off time.

    I'll probably keep using RAW/CRAW burst mode more now. But I now have a better feel for its strengths and limitations.

    I think I need a double single malt after thinking about and writing all that.

    90% luck, 5% field craft, 5% camera skills.

  • AI tracking software will either not keep up with any subject changes

    The Nikon has a tracking "sensitivity" option, which scales 1-5 on interruptions to the subject being in the AF box, ie bird moving behind twig/ bird in flight inconveniently flying behind a tree etc, which changes the amount of time the subject is out of view before the AF re-engages with the surroundings... I'm sure Canon will have some option similar, but again, not familiar with the R7.

  • Are you using Back Button Focus? There was a reasonable chance the bird would go vertical. With back Button focus you would have focused on the bird, lifted your thumb slightly and the bird would have stayed in focus while stationary and probably all the way up as the camera would not have tried to focus until you put your thumb back on the button. The focus on the eye has gone by photo 2 and I suspect the camera is already hunting especially as I suspect you are in servo mode.

    Rolling shutter would not have been an issue in this shots really - rolling shutter usually affects the bottom of the frame as the shutter moves down so the disfigurement occurs at the bottom when it does and its usually when something is moving fast. Try getting someone to golf swing across you while you shoot in electronic shutter and you should get distortion that is totally unnatural. I saw an example the other day of an image where a Water Rail face on to the photographer shook its head and the distortion of head and bill was awful but was not motion blur.

    What shutter speed was used?

  • subject tracking (currently set to recognise animals)

    I had another thought in the night Angus, do you have specific bird recognition rather than animals? Nikon did a firmware update to introduce bird recognition and that made a huge difference over the previous animal recognition.

  • Not using back button focus, but I can see the focus rectangle firmly on the Jay. It is only a blue rectangle. Subject and eye tracking normally has a black rectangle, which is why I suspect the R7 is not tracking when in RAW burst mode - not enough processing power, reduced software capability, etc due to the R7 being a penultimate base model. Standard degradation for price points.

    I was trying to try photographing a Blue tit or Great spotted woodpecker either landing or taking off from my bird feeder. They have very fast wing beats.

    Shutter speed was 1/664.I'm pretty lazy and leave my camera on Program mode. Which is why I suggested Tv (Shutter priority) mode for photos such as these. I can set the shutter speed to be very fast. The iso was 2500 in program mode.

    I'll continue with experiments.

    90% luck, 5% field craft, 5% camera skills.

  • Canon R7 has two sensitivity option: Tracking sensitivity and Accel/Decel tracking. Seems to take values of +/- 0-2. I normally have these values set to -2. This seems to work for my style of photography, and is recommended on various posts by professional photographers.

    For 99.9% of my photographs sensitivity values of 0 to +2 are too fine. The camera hunts like crazy.  Even at -2, the sensitivity is very fine. If either myself or the subject moves the tiniest millimeter then the subject becomes out of focus. Most annoying.

    On the other hand, there have been a number of firmware upgrades since the R7 was released. The way tracking sensitivity operates may have changed. Time for experimentation.

    The Canon R7 has further tracking options.  I have mine set to Case 3 - Instantly focus no subjects suddenly entering AF points.

    Case 2 is Continue to track subjects ignoring possible obstacles.

    Case 1 is generic, probably a combination of 1 and 2.

    Again, how these operate may have changed in the light of firmware upgrades, so further investigation is required.

    90% luck, 5% field craft, 5% camera skills.

  • The R7 offers me the grand choice of People, Animals or Vehicles. I once tried some of the other settings. Doesn't seem to make much difference what the R7 locks onto whichever option I choose.

    With my software engineering hat on, allied with my knowledge of computer hardware and smaller knowledge of AI technology, I can see why any camera would have difficulty recognising the Jay as a bird just as it uncurled to leap upward. Matters were compounded by clutter. So many different edges and shapes to sort out.

    The underlying reason why the R7 isn't tracking is, I feel, because it never was performing eye/animal tracking in the first place while in RAW burst mode. The black rectangles, signifying tracking do not appear.

    I tried this out on a Red Spotted woodpecker. When not using burst mode, the R7 instantly locks onto the bird and its eye; black rectangles around eye. With burst mode, no rectangles.

    I will try further experiments.

    90% luck, 5% field craft, 5% camera skills.

  • Just for a laugh, I thought I'd try and establish what the canon R7 was focused on.

    I used Canon's imaging processing package: DPP4. I only use DPP4 under duress. It's a pants piece of software, that a memory hog, causing my laptop to crash.

    A Great spotted woodpecker with focus point on back of its head.

    Now the Jay.

    Hmmmm. Either RAW burst mode doesn't store focus points or it does so in a different manner such that DPP4 can't extract it.

    I can't speak for other camera manufactures, but Canon's software does come in for a lot of criticism. Bad software specification and design. Been there, done that, had the arguments many a time.

    DPP4 is pretty bad. It tries to load everything into memory, and holds onto several edits at once, compounding the problem. Loading everything into memory, and not paging/swapping is a pretty ancient way of programming. I'm also mystified as to why DPP4 doesn't let memory management units page it.

    Oh well, more seat of the pants investigation and inference.

    90% luck, 5% field craft, 5% camera skills.

  • I've edited this post to remove my utter drivel. Explanation follows Pimpernebloke's post.

    This is an example of the sheer speed at which birds move, precluding any chance of you tracking it. I missed this starling taking off from my bird feeder (as it was so fast) and only managed to press shutter release when it was half way through the semi-circle it was flying to get to the top of the feeder.

    These 14 photos span 1/4 of a second. Yep, 0.25 of a second. A blink of an eye spans 0.1 to 0.4 seconds.

    No way I could track this manually. It took all my concentration to figure out when to press shutter release.

    90% luck, 5% field craft, 5% camera skills.