Edit: to be clear, lossy compression is not an issue per se, the problem is that the lossy compression is worse than the average lossy compression used by other manufacturers. In any case, it would be great to have options for lossless + higher quality lossy.
Yeah. I'd be entirely willing to have a partial tradeoff that had file sizes slightly higher than the current lossy RAW and instead of using a tone curve plus the "strip quantization", only used a tone curve and encoded a few more bits per pixel similar to what Nikon does. Such an approach would be computationally efficient and save space. For example, using a tone curve to take things down to 12 bits/pixel would be almost indistinguishable from lossless in nearly any scenario (most definitely avoiding the artifacts that stripe quantization has)
In fact, with most Sony cameras, just bit-packing would save some space. Sony's lossless RAWs use 16 bits per sample regardless of what the underlying sensor bit depth was.
Agreed. Now that they have enough computing power, I hope Sony allocates enough resources (programmers/engineers) to make such things happen.
Another mode I would love to see in cameras is a variation on continuous shooting called "accumulator mode" - Reserve some RAM that is might higher bit depth than the sensor, and take each shot and add it to that buffer. You could burst multiple shots to memory without filling the flash write buffer much, permitting artificial synthesis of very long shutter speeds while only requiring moderate ND filters at most. For example right now, in bright sunlight you need a 3 or 4 stop ND filter to get an A7M3's burst rate slow enough to have near 100% shutter duty cycle and not back up the buffer. If the camera could accumulate internally, you'd only need a 1-2 stop filter, maybe even less.
Yes, that would be awesome, and those shots would be re-aligned, right? And by the same method, you can also accomplish this:
- reduced noise in low light
- improved detail, you could take a burst of the moon and have the camera create a combined high detail shot.
Doing alignment would require a LOT more processing power.
I would say that alignment just needs to have access to enough memory, and processing power is not really an issue, because it's a step that is performed at the end, after the burst has been taken, so it's not time critical, and it's fine to make the user wait, and it could display such a message: "processing, please wait ... 50% done, 00:03 left".
Also, only a burst of 10 shots is necessary, after about 10 shots, you won't get more details, so that also solves the memory issue somewhat, and also limits the processing time. Also, if the software is well implemented, it can use the gyro data from the camera which then makes alignment very straightforward, as long as your intention is to shoot a static subject.
At least the use case I see is primarily improved DR in cases where a long exposure was already acceptable (e.g. on a tripod) - people do still use tripods!
A technique I use is to do the following:
Put a weaker (3-4 stop ND filter) on the camera
Put camera into continuous drive mode and ISO 100
Set exposure to preserve highlights
Run the camera for a while
Stack them in post
This gives you the "smooth water" effects of an extremely aggressive ND filter, but without the color casts and with much higher dynamic range because you're synthesizing an ultra-low ISO. You also gain the ability to frame your shot.
Understood. So you want the camera to do that internally, right?
The problem is that you need to get the inter-frame time up to around 2-3 seconds if you don't want the buffer to fill up. Internal accumulation would allow you to shoot with exposure times as short as 1/maxframerate without any buffer limits.
Wait, what do you mean by "buffer" here? Do you mean the reserved RAM for stacking the images, or the buffer that stores each individual image?
Let's call the accumulator RAM "stack". If you have it with enough bits, then you don't need to store anything in the buffer anymore. The RAWs will be 14bpp, so their max value (for one pixel) will be 11111111111111 in binary, normally that should be 2^15 - 1 = 32767. If you allocate 15bpp, then you can store 2 accumulate 2 images in the stack, 16bpp => 4 images, okay, so 2^(bpp - 14) images, so if we use 32bpp, we can accumulate 262144 shots and that "stack" buffer would take 232MB (for a 61MP image). If we use 24bpp, then we can accumulate 1024 shots, and that stack buffer would take 174MB.
With 2 small buffers, it's possible to accumulate with virtually no limits.
