DEMO: Dunkirk - Resolution difference between 1080P Blu-Ray and 4K/2160P UHD Blu-Ray (It's CES 2018 also!)

A friend who works in the world of video production saw my post a few weeks back about "4K, Dunkirk and 70mm IMAX" sent me some segments of Dunkirk from the recently released UHD Blu-Ray set so I could have a look at the differences between 4K compared to standard 1080P Blu-Ray. Remember as discussed a few weeks back, this movie was filmed to a large extent with 70mm IMAX (65mm/15perf) gear, we know the VFX rendering was done using >4K with a true 4K digital intermediate. It represents the "pinnacle" of analogue film technology mixed with the best of today's digital effects and processing. I highly doubt we'll see much more of this level of literal "film-making" over the years. As such it's a nice "benchmark" for comparisons.

The segments he sent me were from the ripped discs - the 4K/2160P UHD Blu-Ray version in Rec.2020 "HDR" color space (10-bit HEVC with average video bitrate of ~60Mb/s), and the 2K/1080P Blu-Ray rip was in standard Rec.709 "SDR" gamut (8-bit AVC with average video bitrate ~30Mb/s). Note that I have my own store-bought copy of this UHD Blu-Ray and I'm happy to support UHD Blu-Rays so let's not talk about how the ripping was done (I actually do not know nor particularly care). The purpose of this demonstration is purely for education.

As you know, I've spoken in the past about the importance of color, contrast, dynamic range as likely the most important factors when it comes to clearly noticeable differences between 1080P and 4K/UHD. Remember that the human eye does have limitations with angular resolution so depending on the size of one's screen and how far away we're viewing the image, there's only so much pixel resolution we'll need before the perceptual threshold of visual acuity is surpassed.

For the comparison today, I used the excellent DirectShow renderer madVR to help! You'll know about madVR if you're a home theater PC aficionado. This fantastic free piece of software is able to perform qualitatively some of the best image processing such as upscaling video. I've played with the software off and on over the years, and without question if I viewed movies primarily off a computer in my home theater, this would be the way to go for upscaled playback of 1080P to a 4K screen.

Remember though that as with most things in life, there is a price to pay and for high quality video upscaling, madVR does take quite a bit of processing power whether through one's CPU or the GPU shaders with the more demanding settings. Also relevant for what I'm doing today, madVR now has the ability to convert HDR color space into SDR for playback on my SDR 4K computer monitor. Through the computer, I can grab the image for closer "apples-to-apples" comparisons. If one does not do the conversion, HDR images will look washed out with poor contrast on a monitor or TV incapable of interpreting the color/contrast values. As you probably know or at least guessed, the Dunkirk UHD Blu-Ray is presented in HDR10 which is the "default" HDR standard.

Above are the settings used for converting the HDR to SDR image. The setting of 230 nits display peak brightness was quite close to the look of SDR. It's impossible to be exact given the tweaked color grading which significantly increased contrast when they converted the movie to HDR, but this is a reasonably close estimate based on the calculated mean and median brightness using Adobe Photoshop's histogram window.

For the demo today, I'll be using my workstation PC running the Ryzen 7 1700 CPU with an old AMD Radeon R9 270X graphics card, no color calibration applied for the monitor, with madVR set to this for upscaling of 1080P to 4K:

Notice that I'm using the excellent NGU "Next Generation Upscaling" algorithm. I maximized the settings for chroma sharpness and applied "very high" luma anti-aliasing which also retains excellent sharpness. This looks great to me without any severe artifacts and in fact taxes the graphics card to the point where it will drop frames when played in realtime (since we're examining still frames, this is OK). I have little doubt that the image quality achieved with the software is superior to essentially any TV out there doing upscaling of 1080P to 4K.

I used the current newest version of the software as of December 2017 (madVR 0.92.10), my player software is MPC-HC (1.7.13). I've also got the latest K-Lite Codec Pack (13.7.5) installed. Other than the madVR customizations above, the rest of my set-up was maintained at defaults.

Let's look at stills from a few sections in the movie and compare the 1080P with madVR upscaling to 4K and the exact same frame in 1:1 native 4K from the UHD source with HDR-to-SDR conversion. To the best of my ability, I selected frames where there wasn't too much motion blurring and where it appeared that the camera focus was spot-on. As usual, if one "pixel-peeps" too much, it becomes evident just how often scenes are shot slightly off-focus (this can be very obvious with 4K material!).

Remember you must click on the A/B images to view or download the large 1:1 original size PNG lossless images to properly compare! I'll also include a link to a ZIP file with these comparison images below for those who want to download them all in one shot.

Comparison I: Troops on beach (00:06:27)

Many people have commented on how with the 70mm film, we can appreciate the number of soldiers on the beach better. More life-like, more like a "window" into the scene rather than an artificial facsimile.

Sure, we see more definition in the troops. Notice in the distance near the top right, we can see in the 4K version a guy holding his gun propped up on the sand - we only see a hint of the gun in the 1080P upscale. Billows of smoke and the flames are also clearly better with the 4K in the distance. You can also count the number of heads among the mass of soldiers in the far distance with the 4K version as opposed to vaguely discriminant blobs in 1080P.

Comparison II: Sand blast (00:07:11)

A great scene near the start with the dive bomber:

More defined pores and callouses on the knuckles (even though not fully in focus). Fine details of the sand easy to spot. Notice the sharpness of the individual strands of hair in 4K.

Comparison III: Water / Ships (01:26:28)

Nice panoramic shot of the sea, beach, and boats. Despite a bit of atmospheric haze, we can still pretty well make out the number of people on the small boats in 4K compared to the 1080P upscale which again looks blob-like. Also notice the waves in the water which in the 4K version is able to convey the specular shimmer in significantly greater detail. A few of the boats have what looks like red flags or markings better conveyed in 4K. Look to the top of the crop and notice what appears to be some pier/mole railings in the 4K version which is poorly defined in 1080P. Also the details of the structures and buildings on the beach and that docked ship in the top right of the crop looks much improved in 4K.

Comparison IV: Plane (01:30:11)

I liked the plane and dog-fighting scenes. Not sure about the realism of this "gliding" scene in the end though. Really, can you glide a Spitfire for many minutes like that without any power to the engine!?

Nice shot of the Spitfire! Notice the extra details of the pilot and cockpit. The rivets and exhaust behind the propeller are much better defined and "solid" looking. I also like the extra definition on the seams of the plane. Improved color definition of the RAF "roundel" with 4K.

Comparison V: Helmets on beach (01:38:30)

While Dunkirk is far from a brightly lit film overall, let's check out a dark scene:

Of all the samples, probably the most difficult one to make out significant improvements in image quality. Overall a bit more clarity to those helmets and like the scene above, lumps of sand are significantly more defined. In the distance near the water's edge, we can make out individual lighter colored rocks in 4K whereas it's blurred in 1080P.

Remember that the 4K HDR image has 10-bit color depth versus the usual 8-bit Blu-Ray. On a screen capable of 10-bit rendition, details in these darker scenes may be better demonstrated.

Comparison VI: Face close-up (01:38:54)

Notice that whereas the above samples are in 16:9 aspect ratio from the IMAX 70mm (65mm/15perf) footage, this close-up/dialogue scene is presented in letterboxed 2.2:1, shot in 65mm/5perf (see here for comparison and what this means physically).

By the way, remember that Tarantino's Hateful Eight (2015) was filmed similarly in this "lower resolution" 65mm/5perf but with 1.25x anamorphic lenses and expanded to an even wider 2.75:1 aspect ratio! I think that film deserves a 4K transfer to improve the height resolution given how it has to be so significantly letterboxed for a typical 16:9 home theater display.

Again, clearly the detail is better preserved in the 4K facial shot. Pores, lines on the lips, individual eye-lashes, that small abrasion on the tip of the nose (kudos to the make-up artists for details like this!). Notice also the retained definition of the right side of the image where the face is shadowed.

Conclusions

Well, indeed the 4K UHD Blu-Ray represents an improvement over the 1080P standard Blu-Ray as expected resolution-wise. Even with the excellent madVR upsampling of the 1080P material, sophisticated upscaling algorithms obviously cannot make up for all the extra detail that 4K can offer. When information is lost, there's no way to recover it completely (remember this, audiophiles, if you're arguing for MQA!).

If you have a 4K TV connected to the internet, make sure to have a look at the images above at 1:1 resolution sitting at your usual viewing distance and see if you can spot the resolution improvement 4K makes. Obviously if you can't, then it's not the resolution difference that you'll benefit from, so better make sure you like the HDR color/contrast effect before investing in 4K UHD Blu-Rays and only pick the ones offering good HDR enhancement.

Although there is obviously demonstrable improvement, I don't expect everyone to just run out and buy UHD Blu-Rays even if one can see the difference! The 1080P Blu-Ray image looks stunning already upscaled when I watch the movie on my 75" Vizio P75 TV from about 9 to 10 feet away. Yes, I can see the difference in resolution but it's far from earth-shaking. Mr. Nolan might disagree, but the high dynamic range effect makes this the "definitive" version qualitatively compared to the 70mm IMAX film projection I watched for the various reasons cited previously.

Tangentially, one of the great benefits of bringing a movie like this home is that I can adjust the sound volume to my liking. The soundtrack was so loud at the theater that I could barely hear the dialogue (the other movie I'm looking to turn down the volume on is Blade Runner 2049 coming soon)! If you're a Nolan fan, do check out the 4K Collection. IMO a good deal if you like the movies.

Remember that there are all kinds of other factors involved in image quality comparisons. Video is moving so while still-frames can show a difference, minute differences might not be as noticeable or evident during playback. UHD Blu-Ray is also 4:2:0 chroma subsampled like Blu-Ray which implies some loss in color resolution. Another variable is that bitrate might not be optimized; this should not be an issue here since the movie is only 1 hour 45 minutes long and would fit on the disc with no need to compromise bits. Bitrate however is often why streaming quality isn't up to par with an actual disc capable of delivering up to ~100Mbps off a UHD Blu-Ray if needed (compared to Netflix 4K with something like up to 25-30Mbps including the audio data stream). Remember also that video is encoded in a "lossy" format whether it be Blu-Ray's VC-1/AVC or UHD Blu-Ray's HEVC so image quality can vary depending on the encoder. Finally, specifically for this demonstration, remember that I'm using selected madVR settings for upscaling and HDR-to-SDR conversion; the settings might not be exactly optimal for my set-up and this film but I highly doubt the upsampling can look much better!

Despite the improvements we already see between 4K/2160P over 2K/1080P in this film, there's still a lot of resolution to be tapped with 4K. I suspect we'll have to wait for something like a Pixar CGI film rendered in 4K/HDR with vivid colors to better appreciate what the technology can do - Pixar's recent Coco for example is still a 2K render. Also remember that UHD Blu-Ray can handle 50/60fps; imagine a pristine animation with high framerate and lower motion blur. Recently I watched the 4K version of Your Name. (Kimi no na wa.) (2016) for a tantalizing taste of just how impressive "ultra" high resolution computer-assisted animation can look!

Download ZIP file with all the Dunkirk comparison images here.

-------------------------------

This past week was CES 2018. Good to browse the news items on The Verge, AnandTech even PC World every year to see what's hot and what catches the eyes of the editors there.

In the audio world, I'd be interested to hear Sennheiser's closed-back HD 820 headphone featuring Gorilla Glass. Creative looks like it's doing HRTF modeling with Super X-Fi 3D audio through headphones to render multichannel sound; this could be a great way to kickstart the production of more multichannel audio if there is demand from the head-fi masses! IMO, the multichannel experience is qualitatively a major upgrade when done right. There is potential to move forward if one can get past the cost issues and space demands by using headphones. One concern might be safety issues if the 3D surround sound seems too realistic and absorbing in some settings :-).

Obviously 8K TV is a questionable idea for the time being (and I suspect for the foreseeable future when it comes to content despite the push from NHK).

Notice by the way the lack of coverage of hi-res audio in the mainstream. And of course the yearly lament that "high-end audio" is no longer represented at the CES tradeshow. Remember that CES focuses on innovation and trends for the upcoming year. When it comes to audiophilia and typically 2-channel sound, what great advancement do we expect from a very much mature technology? Yet more expensive Technics turntables? Is there truly innovation in expensive cables (magnetic cables? monocrystal "rare earth"!?)? At least back in 2015, MQA was given a fair shot of PR alongside Neil Young and the PonoPlayer. Time marches on and we obviously know more about these "technologies" and the (lack of) impact they make. These days the "high end" is about luxury, not innovation. Sure, audiophile brands should exhibit to some of those shows catered to direct contact with the public who are specifically interested in buying such specialty items. The disappearance of "high end audio" from CES is not unexpected and perhaps necessary to reflect the role and audience CES appears to be targeting.

Here's an interesting claim in the CES news cycle that hi-res audio devices are selling and market share is growing (data originally released in fall 2017). Check this out from NPD Group:

What's the big deal? Of course there's "growth" when standard AV receivers can do 24/96+ these days as just part of basic functionality since any decent commodity DAC chip supports >16/44. A full 83% is due to receiver/amp sales and soundbars. And how does one define what headphones are "high-res", or are they all capable of high-res? Whatever, guys... As I said last year about CES and the push for high resolution audio, please focus on releasing some decent music that can take advantage of higher resolution. Hardware adoption numbers like this seem rather meaningless unless we're clear about the methodology. Consumers aren't stupid nor does this data seem all that impressive when there's still barely any worthwhile high-res music out there as there is no actual initiative to call out the elephant in the room by the Industry and their mouthpieces to make changes to ensure better music production.

I had a listen to the recent Gregory Porter album Nat "King" Cole & Me (2017, nice DR13) - good set of Cole covers if you like "standards" vocals. Don't bother with the 24/96 high-res download:

Yet another example of "pseudo-high-resolution". At best this is a 16/48 track. Just a bunch of high frequency noise essentially uncorrelated with the music up top. Looks like they ran the 48kHz digital through an analogue console and recorded it back as 96kHz! Why Blue Note, why do you do this?! If my suspicion is correct, they should have just released this in 24/48 and possibly retained a lower noise floor... Certainly no shame in 24/48 high resolution!

Hope you're all enjoying your music and movies...

Have a great week ahead!