Thomas Worth recently achieved the holy grail of Cinema DNG compatibility for ML RAW files with the released of his fantastic little converter app RAWMagic. Unfortunately a fair few people had been reporting on the forums that it was not quite the revelation hoped for. I thought I’d try and put it to the test, Resolve against the old workhorse, Adobe Camera RAW.
These are not the best controlled or most scientific tests you’ll ever see. But I think they’re good enough to draw a few conclusions. I’ve tried to avoid using any of the ‘toys’ available in either application. All sharpening and noise reduction was disabled and only white balance and tint controls adjusted (plus EV in the noise tests). The aim being to compare the characteristic inherent in the debayer algorithms and colour space conversions only, not to compare the applications themselves. This was never going to be entirely possible, but I’ve done my best.
For these tests I used a Canon 5D mk II with a Canon 85-mm f/1.8 lens. The camera was running a.d.s 6th June build (d5a606e8c616) of the ML firmware and recording 1880x1056@25fps. The .raws were converted to dng with RAWMagic Beta 3. ACR 7.1 (AE 11.0) and Resolve 9.1.3 were then used to process the images to 1080p25, 10-bit Uncompressed YUV 422 files in Rec 709 (data levels for easy online display). All comparisons were then done using these files, to at least vaguely mirror a realistic workflow.
Download the full res images to go with this article:
Sharpness and Aliasing
The first image captured was of a fairly standard lens test chart. It was lit with two 150W tungsten lamps covered with Lee Filters 3/4 CTB gel. As well as showing some fairly striking differences in approach to handling detail/aliasing, it showed up a discrepancy between white balance and tint settings in the two apps. These two files were converted with the same values for these parameters.
1/25 sec @ f/4 ISO 400
From this image I’d say ACR suppresses the aliasing more effectively, especially the chromatic aliasing, but this is not universally the case and it is quite noticeably less sharp than Resolve. However, looking at less abstract imagery the difference seems even more pronounced. Look at the vertical scale on the left of the colour checker chart below. And in the real world footage at the end of this article there’s some even more striking effects.
Colour and Gamma
This was shot in the same set-up as before but using a pocket colour checker and some paper money as subjects. Again WB and tint were adjusted, but this time attempting to compensate for the difference in the scales.
1/50 sec @ f/4 ISO 400
Resolve has used a steeper gamma but apart from that colours are pretty similar in the mid tones and highlights. Reds and yellows seem a little more saturated in ACR but this could be experimental error. However, the shadows have a distinct magenta tinge from Resolve that is not there at all from ACR.
It’s quite possible for a debayer algorithm to suppress, enhance or even add to the noise in an image. To test how the processes differ on this count I used the same set-up as above, but this time under exposing by 3 stops and adding the gain back in through camera raw.
1/50 sec @ f/11 ISO 400 (+3EV in post)
ACR renders an almost monochromatic noise that is less visually abhorrent than the more colourful version from Resolve. It does this at very little cost in terms of colour resolution, at least not any thats visible in these 422 files. How it does this I don’t know, but it’s very effective.
I also repeated these test under tungsten lighting but I could see no significant difference in how the two systems handled this so I’ll say no more.
To round things up I took three pieces of footage that had shown up problems when processed through raw2dng and ACR, and re-processed them through the same two channels as above. See for yourself (download it, don’t watch on vimeo!). Look out for colour aliasing and harshness in the bluebells and the water surface, and the colour of the in-focus highlight from the moonlight shot.
ACR was designed when 2 MPix chips with less than 30% light sensitive surface and no low pass filter were state of the art. DaVinci’s RAW engine was born in the era of 20 MPix sensors, micro-lenses and sophisticated anti-aliasing filters. It’s no wonder ACR is better suited to the needs of ML RAW. How much better suited is perhaps a little surprising.
I will definitely be using RAWMagic to convert my RAWs, and Resolve to produce proxies directly from the cineDNGs. It’s worth pointing out the Resolve did all its renders at least 10 times quicker than AE, and I’m not even running it on it’s preferred architecture. But for final grade I’ll be going back to ACR and re-rendering the clips I need. For all its disadvantages it does hide a multitude of sins.