Denoising in-camera multiple exposures RAWs

(Pat91) #1

Hi,

A friend of mine has tried to denoise RAWs produced by using the in-camera multiple exposures mechanism of an EOS 5D MK III. Both HQ and Prime in DPL 2 are totally unable to denoise such images. Acceptable results are obtained with Lightroom and excellent results with the Imagenomic Noiseware plugin used in PS or Affinity Photo.

Is this actually expected ? Did I miss something or is there a problem with these particular files ? I can upload a few of them if needed.

Thanks in advance.

1 Like
(Svetlana G.) #2

Hello @Pat91,

Yes, please, provide us with some samples.

@wolf could you, please have a look?

Thank you,
Regards,
Svetlana G.

(Pat91) #3

Hi Svetlana,

I sent you a private message with a link to some sample files. My friend doesn’t want to make these images public.

Regards.

(Svetlana G.) #4

Thank you. Got them and provide to wolf for the investigation.

Regards,
Svetlana G.

(Pat91) #5

Hi,

Still waiting for an answer after almost 3 months. I’m really interested in a technical explanation about this issue. My friend is working on a large series using the multiple exposures technique and was forced to drop DPL for this task.

Thanks.

(Svetlana G.) #6

Hello @Pat91,

Sorry for the delay. Actually, it’s not an easy issue to fix - the images were analyzed and the reason found but it needs time to fix it.

Regards,
Svetlana G.

(Pat91) #7

Hi,

I can understand that there are technical difficulties with this and that fixing this problem is not at the top of your todo list. No problem. Just, we’re interested in the “why”. And we’d like to know whether it’s a definitive “no” (like for the Fuji X-Trans files) or just something that was overseen and will be fixed one of these days.

1 Like
(Alec Kinnear) #8

I’ve had some success providing finished jpegs to PhotoLab and using Fast Noise Reduction on those jpegs. Fast Noise Reduction can have a pleasant graininess which reminds one in a good way of analogue noise.

Your question is a good one: I’m keen to see support for at least Iridient Transformer (demosaiced) Fuji X files.

(Pat91) #9

Bump !

You said the reason was found. Even if the problem is difficult to fix, we’d like to know the “why”… Again, you can get technical, I’m a former developer.

(Svetlana G.) #10

Hello @Pat91,

It’s a question not to me but @wolf. Let’s ask him if he can clarify the reason.

Regards,
Svetlana G.

(Pat91) #11

Hi Svetlana,

I had understood that you already asked @wolf to reply our question but since he/she never posted in this thread… Thanks anyway.

2 Likes
(Wolf) #12

Hi Pat91,

We have investigated and understood the issue, already back in April. But, you know, we’re a small team and can’t handle every issue instantly… Good news is that we’re right now working on this topic, so stay tuned.

Br, Wolf

1 Like
(Pat91) #13

Hi Wolf,

We have already understood this but that was not the question. We are curious about the cause of this issue. We understand that it may take some time to fix it but we are interested about a brief technical description of the problem. Me and some others here are skilled enough to understand a technical explanation in case this is something specific to the files containing multi exposure images. If this is merely a bug, we’ll be patient.

1 Like
(Wolf) #14

Hi Pat91,

Ok I got it. Here you go:

We calibrate the noise model for each camera model in our lab, in function of ISO. In normal shooting conditions, all cameras of the same model will produce the same amount of noise, when shooting at the same ISO. So we can use the calibrated values as parameters to our denoising algorithms.

It turns out that, in multiple exposure mode, the noise differs significantly from the calibrated value (probably because the read-out noise accumulates, but maybe there are some other phenomenons that only Canon themselves could explain). Therefore our denoising uses parameters that are not well adapted to the actual noise level, and does not work well.

So it’s not a simple bug – it’s a limitation of our assumption that the noise model can be calibrated in the lab.

Br, Wolf

6 Likes
(Pat91) #15

OK. Thanks for the explanation.

(Jim Scott) #16

Ditto - Thank You for the explanation!