Fujifilm EXR "DR Mode" raw files

floessie · May 6, 2019, 2:09pm

Just FYI: For DR (“dynamic range”) RAFs you can use HDRMerge. @heckflosse made a change that enables this. Don’t know for X-trans, though.

HTH,
Flössie

Nando22 · May 6, 2019, 7:41pm

The RAF file has 2 images. You say that HDRMerge, merge both images? How can you separate the two images in HDRMerge to indicate the difference in exposure?

heckflosse · May 6, 2019, 7:46pm

HDRMerge does this automatically if it finds 2 raw frames in a raw file. Maybe start hdrmerge -vv (two v, not one w) from console to see in console output if it works correctly.

Sunhillow · May 6, 2019, 9:02pm

I guess you are talking about either EXR or SR Sensors from Fuji?
X-Trans only contain one image

Nando22 · May 6, 2019, 10:19pm

Yes. Fuji X10 sensor

heckflosse · May 6, 2019, 11:17pm

Fuji X10 has EXR-sensor. If the raw has two frames with different exposure (DR mode), HDRMerge should be able to combine them. If the raw has to frames with equal exposure, use Rawtherapee with SN mode
grafik

Nando22 · May 7, 2019, 11:27pm

It didn’t function. Perhaps I don’t know how to work with HDRMerge. I use Luminance. Do you prefer HDMerge over Luminance for some reason?

heckflosse · May 7, 2019, 11:35pm

Can you provide a raw file to test?

afre · May 7, 2019, 11:45pm

They do different things. One merges raw files into a single raw file in the interest of noise reduction; the other merges raw files, tone maps the result and outputs a processed image.

Nando22 · May 8, 2019, 12:01am

DSCF7760.RAF (18.9 MB)

Nando22 · May 8, 2019, 12:08am

This is my test file. My workflow is:
1- load in RT
2- generate tiff with frame 1
3- generate tiff with frame 2
4- load in Luminance
5- merge with 2EV difference
6- generate tiff 32 bits
7- load in RT
8- process (tone mapping in RT)

Nando22 · May 8, 2019, 12:09am

And this is the final result

Morgan_Hardwood · May 8, 2019, 9:32am

This is from HDRMerge (branch cmake-exiv2 commit 9d0050d), merged using a white level of 3700, then processed only in RawTherapee 5.6 (no need for LHDR):
DSCF7760-7760_wl3700.dng (10.9 MB)

DSCF7760-7760_wl3700.jpg.out.pp3 (11.1 KB)

floessie · May 8, 2019, 9:44am

Very nice and natural. (I expected nothing else from you. )

I gave it a try and ran into the white level issue as well. How did you come up with 3700?

Morgan_Hardwood · May 8, 2019, 10:23am

I measured the RAF file and was surprised to find a buildup of values starting from around 3700. I expected to see a spike, not a bunch.

As I don’t have time to look into this, I hope someone else can shed some light on why there is a bunch and not a spike, and also on what exactly it is that I’m looking at - is it a histogram of frame 1 or 2, or a blend of both?

Morgan_Hardwood · May 8, 2019, 11:11am

3 posts were merged into an existing topic: Fujifilm EXR “SN Mode” raw files

Nando22 · May 8, 2019, 10:55am

Thank you. I have to study how to use HDRMerge. Are any good tutorial to recommend?

Corwin_Black · May 8, 2019, 1:19pm

Can I get sub-image or similar option for Fuji S5 Pro? Sure, Adobe RAW does really good job these days with it, but still.

heckflosse · May 8, 2019, 1:32pm

No. The rt engine does not support 2 frames of different dimensions atm.

Edit: for reference:

github.com

Beep6581/RawTherapee/blob/dev/rtengine/rawimagesource.cc#L1577

    
      
              }
          
          
    std::unique_ptr<CFALineDenoiseRowBlender> line_denoise_rowblender;
              if (raw.bayersensor.linenoiseDirection == RAWParams::BayerSensor::LineNoiseDirection::PDAF_LINES) {
                  PDAFLinesFilter f(ri);
                  line_denoise_rowblender = f.lineDenoiseRowBlender();
              } else {
                  line_denoise_rowblender.reset(new CFALineDenoiseRowBlender());
              }
          
          
    cfa_linedn(0.00002 * (raw.bayersensor.linenoise), int(raw.bayersensor.linenoiseDirection) & int(RAWParams::BayerSensor::LineNoiseDirection::VERTICAL), int(raw.bayersensor.linenoiseDirection) & int(RAWParams::BayerSensor::LineNoiseDirection::HORIZONTAL), *line_denoise_rowblender);
          }
          
          
if ((raw.ca_autocorrect || std::fabs(raw.cared) > 0.001 || std::fabs(raw.cablue) > 0.001) && ri->getSensorType() == ST_BAYER) { // Auto CA correction disabled for X-Trans, for now...
              if (plistener) {
                  plistener->setProgressStr ("PROGRESSBAR_RAWCACORR");
                  plistener->setProgress (0.0);
              }
              if (numFrames == 4) {
                  double fitParams[64];
                  float *buffer = CA_correct_RT(raw.ca_autocorrect, raw.caautoiterations, raw.cared, raw.cablue, raw.ca_avoidcolourshift, *rawDataFrames[0], fitParams, false, true, nullptr, false, options.chunkSizeCA, options.measure);

Corwin_Black · May 8, 2019, 5:33pm

Process individually - upsize/downsize/crop - merge?

That highlight part has pretty much always same processing, it just needs to have cutoff at certain EV to not get pink/red highlights and some heavy NR.

Another (simple) solution would be just to split file and let users process and output both, I did that with another SW, results are okay-ish (could be better, if original SW had some reasonable demosaicing and NR).