Short video shot raw, processed/color graded in Rawtherapee

Hello, I am in a student film production club and we shot a short film with a crew of 10-15 people, with properly defined production roles. I was the camera operator, sound mixer, and color grader. I shot the film with a Blackmagic Pocket Cinema Camera, and a set of three russian s16 lenses, MIR-11M 12mm F2, VEGA7-1 20mm F2, and TAIR-41M 50mm F2.

We decided to shoot everything in the bottom of the stairwell of a concrete building to get a decent prison set. Unfortunately, we had to reschedule, and so the second chance, we didn’t have access to serious studio lights, and proper audio recording gear, so we had to make due with in camera mics for diologue, and limited lamps and flashlights to light everything. So we ended up with pretty low light footage. As I will later explain, Rawtherapee helped me rescue our shoot.

For editing, I started out by using Davinci Resolve to transcode all the cinema DNG raw sequences into low resolution h.264 proxy files, and handed them off to another team member to cut and assemble onto a Premiere Pro timeline, as well as select sound effects. They also had the actors overdub and sync the audio. We met and they passed me the files and I exported the timeline to finish in Resolve. In the meeting, I showed the director a demo grade of a still frame from our project to get a general sense of what she wanted it to look like.

To start, I graded a middle frame from the first shot we did (wide angle of the whole scene, for the whole video), and exported it to a separate folder, so that I could use it as a color reference for everything else, then copy/pasted the processing profile across all the frames in the clip and sent to queue. After finishing, I deleted all the raw files and .pp3 sidechain files, so sorry, I have nothing for y’all to play around with (85 gigabytes was a lot of space on my hard drive!). Found out I do have some raw stills and a sidechain file. See reply below for link.

I had to push the exposure in RT about 4.5 stops, (effective ISO 20,000 for an ISO 800 native camera), and from there I drastically lowered the color temperature, decreased the saturation and added contrast, tweaked the tones with a filmic curve, boosted the greens, and desaturated the highlights in the Lab color module. I used LMMSE demosaicing because of the high noise (not possible in Resolve/Premiere), and hot pixel filter because the BMPCC has a crappy sensor with lots of hot pixels, which the RT hot pixel filter did beautifully without artifact (Not Possible in Resolve/Premiere). Another gift the benevolent sensor gave me was a lot of line noise artifacts, which a setting of 3-5 on RT line noise filter got rid of nearly all of the line noise artifacts (Not possible in Resolve/Premiere and has been a common complaint of the BMPCC that Rawtherapee solved effortlessly). For sharpening I did RL deconvolution with a high contrast mask threshold, which looks much better than the old and I’d say obsolete dampening filter, which gave me good sharpness in details, with a more natural rolloff into not affecting the flat noisy areas than would be possible in Resolve/Premiere. I then used impulse noise reduction to even out the noise better, and then did Lab noise reduction, lowered the gamma to make the noise reduction work harder on the noisy shadows an midtones, used the lightest chroma noise reduction settings necessary to do the job, set the luminance NR to slider at 100%, and dragged the detail recovery up until I started seeing noise. I then turned the luminance NR to curve and brought the level in the highlights to zero, and the blacks to 15-45%, depending on how much exposure gain was used in each shot. I was able to seemlessly recover most of the upper mid frequency/ lower mid amplitude detail that was eroded by noise reduction but still had good signal to noise ratio with the edge sharpness wavelet feature. The new auto chromatic aberration correction in RT 5.5 did an excellent job correcting the excessive chromatic aberration in the wide angle Russian s16 lens (MIR-11M 12mm F2). Manual vignetting and distortion correction worked in a pinch, but I suppose I should stop being lazy and make lens correction profiles for my Russian s16 lens kit.

It took my modestly powerful editing PC (AMD - Ryzen 5 1500X 3.5 GHz Quad-Core, 16 Gb of RAM, and Seagate BarraCuda 1 TB 3.5") about eight hours to process the 45,000 frames, and more time on that doing the actual editing, but it was worth it. The overall result was, Rawtherapee enabled me to turn heavily underexposed footage from an outdated and small (less than µ4/3) sensor, into modestly grainy, but organic and film like looking footage, not clean but pretty good as a film aesthetic. It just looks so good to me, compared to what I see on Youtube of people doing low light tests with the BMPCC.

The video is viewable as an unlisted Youtube video linked below, and as a google drive link to the original high quality final rendered file below. Note: to actually view the original file in high quality, download the file in the google drive folder, do not just use the online video preview function.

Google Drive link:
https://drive.google.com/open?id=1xVxpNCH9MySKy1xOv9_sDt6QRlig5xxy

Youtube link

One thing that could be improved would be a scale to fill crop feature in Rawtherapee. For correcting barrel distortion, using raw edge pixels is helpful in reducing the amount of scaling needed, but in setting the raw border to the intended standard video resolution output, the edge pixels are not used when performing the distortion correction. If I use the composition tools to to fix this, I end up having to have the image unnecessarily re-sampled a second time, once to do the distortion correction, and another to get the proper image size. This could be solved with A, allowing the user to set the auto scaling in the transform tab to the crop dimensions, or B, have the edge pixels shaved off in the image border in the Raw Border setting be processed and available for use in the Transform tab.

Another feature that would be helpful for video work (correct me if this is already available), is to be able to assign processing profiles to folders, so that they apply to all the still frames within said folders, editable by editing any file in that folder. Also sending folders and included images in said folders to queue without loading all the previews to RAM would make the workflow more manageable.

TLDR version: I helped make a student video (linked above), and used Rawtherapee to raw process/color grade. Workflow was cumbersome, but the results were incredible.

Edit: Turns out I do have some sample raw files and a .pp3 side chain file. See post below.

13 Likes

I agree that the dampening filter is obsolete since the introduction of the contrast mask. I kept it for backwards compatibility with old pp3 files. Maybe we should remove it in a next version of rt to get a cleaner interface. Just an idea…

Edit: still downloading your video to see it in full quality…

2 Likes

Good work.

A note on sound mixing: when music is behind dialogue, it can be overpowering when the music is similar to human speech, eg violins and cellos. A useful trick is to reduce the music amplitude in the range of human speech, eg 80 to 260 Hz.

2 Likes

Any chance you have a torrent file?

Nice testimony, thanks!

Is it possible to code Rawtherapee to interpret old pp3 files with dampening settings as contrast mask settings that produce similar results?

Edit: Also, it would be cool to be able to gamma weight (bias towards targeting shadows or highlights with a basic gamma slider) the contrast mask for both demosaicing and sharpening, like can already be done in RT with noise reduction.

I actually thought of the conflict (also not my decision to have music going the entire time). I actually cut the music at the frequencies in the 1kHz to 4kHz zone (speech intelligibility region), during the dialogue. Not sure how reducing music around the fundamental frequency of the human voice (not as much intelligibility information ) would help. Care to elaborate your thinking?

I don’t even know what a torrent file was until I googled it. Lol.

Thanks for sharing!

I agree that the sound design and engineering could improve. I wouldn’t tamper with the background music too much. Instead I would enhance the voices by giving them more definition (fuller) and character (room echo). Also, the overdubbing seems to be slightly out of sync. To me, that is the most distracting element.

I didn’t do the overdubbing, and sound design was very rushed. I did give the voices a little bit of room echo, though, was worried about overdoing it.

I understand and I guess you didn’t have much time to edit the audio either.

I am kind of more curious on your opinion of the raw video processing. Both aesthetic and workflow. I kind of expected that to be the discussion, especially since raw processing and color grading video in Rawtherapee is uncommon.

1 Like

It’s just a question of fiddling with graphic equalizers and faders until speech can be easily distinguished from the music, and we can “block out” the conscious music to hear the speech.

If the audience has to strain to make out the words, there is a problem. This often happens with amateur videos, rarely with professional producions.

The high-quality video is more than 4GB, too large for me to download, sorry. The aesthetics of the color-grading look fine to me. I’m less happy about the camera angles. The traditional way (start wide, with each successive shot closing up on the faces) seems more suitable than interspersing some random wide shots. I did like the close-ups on the hands.

The main thing is that we don’t have original footage or frames to compare against, so the amount of feedback we can provide is limited. Unfortunately, I don’t have much experience with video and colour grading. I will try to give some feedback:

  • I would say that the cuts are a bit rough. I would have liked there to be more breathing room between cuts. I mean, they aren’t arguing and the scene isn’t frantic.

  • Colour grade more aggressively to suit the mood and have it transition as the mood changes toward the end. Below is an example from Daredevil. (I don’t recall it being so hazy / low contrast. Maybe the uploader did something.) The oranges and reds really add to the tension and danger, and likewise the blue-green of the hallway tell another story.

  • The shirt and pants of the woman to the right are blown and crushed. The shirt needs to be brought down and the blacks a little lift. The shirt in particular doesn’t match the lighting of the rest of the scene and is distracting.

image

I don’t think that’s possible

Dang, that sucks. Seems pretty basic.

Will you be doing more of this?

Yes. The quality I get out of RT processed raw video is breathtaking. Now that I think of it, I might have the some raws on an SD card. Will upload later tonight, so we can discuss more technical specifics of the process (what I expected to be discussed here).

Turns out I do have some raw stills. 1st a still from the wide angle take.
Blackmagic Pocket Cinema Camera_1_2019-10-16_0709_C0001_005120.dng (1.8 MB)

Next, a still from the water dropping sequence and corresponding processing profile. Note that the Rawtherapee noise reduction settings were a bit more aggressive on average compared to the side chain file presented below, as I was able to light this shot properly, unlike the majority of the shoot.
Blackmagic Pocket Cinema Camera_1_2019-10-27_1847_C0006_000000.dng (1.8 MB)
Blackmagic Pocket Cinema Camera_1_2019-10-27_1847_C0006_000000.dng.pp3 (11.9 KB)

1 Like

Why did you use LMMSE demosaic with 6 enhancement steps instead of IGV demosaic? Just being curious because (especially for your kind of processing) IGV is on par with LMMSE quality wise, but way faster.

I always thought http://rawpedia.rawtherapee.com/Demosaicing could use some fleshing out. E.g., LMMSE and IGV are just lumped together. Incorporating this tip might help and how it is “on par”.