Picture Quality Evaluation

It’s me again :wink:

As promised I wanted to evaluate the PQ of the ATV and of infuse. I dont have much time right now so this is just a small and quick overview.

Display device is my JVC X35 on a 105" screen.
Connected via AVR with all video processing disabled.
I tested the ATV with infuse, MrMC (hardware acceleration VideoToolbox and AVFoundation is enabled) and Plex (because they use the stock player) and the reference device is a Popcorn Hour A-500.
Note:
Projector was calibrated but since has run 2,000 hours so that its not calibrated anymore.
Differences in color/brightness are due to overlays when the video is paused in certain apps.

Chroma upsampling:
PCH:
https://postimg.cc/image/4pp1qt81h/
Infuse:
https://postimg.cc/image/n59io7tvp/
MrMC:
https://postimg.cc/image/dkpw1d74l/
Plex
https://postimg.cc/image/7js74as85/

Infuse has an excellent Chroma upsampling, even better than the PCH.
Ranking:
Infuse
PCH
MrMC
Plex (worst by far as they only use the simple nearest neighbor method)

PCH:
https://postimg.cc/image/ijdefxl85/
Infuse:
https://postimg.cc/image/64qmfm95h/
MrMC:
https://postimg.cc/image/64qmfmwat/
Plex:
https://postimg.cc/image/iw4sm4yd1/
PCH and Infuse very similar but I give the edge to Infuse. Then MrMC and Plex.

Clipping:
PCH:
https://postimg.cc/image/fpa92ftbp/
Infuse:
https://postimg.cc/image/mfqqbv8rp/
MrMC:
https://postimg.cc/image/8ytrszqqd/
Plex:
https://postimg.cc/image/iw4sm1b6t/
All but Infuse pass the clipping test. This is something you definitely need to have a look at.

Banding (all four dots should be visible in all boxes):
PCH:
https://postimg.cc/image/msi4i391x/
Infuse (sorry its not sharp, but there is no banding):
https://postimg.cc/image/w0acysntx/
MrMC:
https://postimg.cc/image/4pp1qv5hh/
Plex:
https://postimg.cc/image/5f7u390w5/
PCH and Infuse pass as they dont exhibit banding. However, MrMC and Plex show clear banding.

Colorspace conversion:
PCH:
https://postimg.cc/image/lrhvsfx1x/
Infuse:
https://postimg.cc/image/7l24x81md/
MrMC:
https://postimg.cc/image/dyr80gysl/
Plex:
https://postimg.cc/image/3oot18ocl/
The same result as for the clipping: All but Infuse pass. If you would have applied the wrong BT.601 conversion instead of BT.709 there should be dots inside the red and white boxes. However, there aren’t any. I dont know if this is due to my projector not being properly calibrated anymore of if there is something else going on. So please investigate here as well.
That said, all pictures regarding the colorspace or banding should look identical to the PCH as this was tested to have zero errors in this regard.

Upscaling via 720p AVC mp4:
PCH:
https://postimg.cc/image/cjpnbs0ad/
Infuse:
https://postimg.cc/image/s56yvqro5/
MrMC:
https://postimg.cc/image/9cv3s584l/
Plex:
https://postimg.cc/image/b4o2n26x1/
Actually hard to decide which is better. Will need to test other patterns.

Some “real life” chroma upsamling patterns in AVC mkv.
PCH:
https://postimg.cc/image/g3bl1lxv9/
Infuse:
https://postimg.cc/image/t7h5eaxmt/
MrMC:
https://postimg.cc/image/z8eubd4th/
Plex:
https://postimg.cc/image/h5lrk5w45/
Hard to rank them:

  1. MrMC. Softer but less pixelated
  2. Infuse
  3. PCH
  4. Plex

I’m under the impression that Infuse does not decode AVC videos but just passes them to the decoder. Therefore, the video should look identical to the internal player which Plex relies on. However, this is clearly not the case. So my question is why? Is it because apple does not support mkv natively and therefore its decoded by software?
Further conclusions:
Infuse offers excellent chrima upsampling but has problems with clipping and correct colorspace.
Plex (internal player) offers the worst PQ by far.

//edit:
It appears that your forum software doesnt allow for pictures inside the thread.
Have a look at the complete gallery here:
https://postimg.cc/gallery/d8291pae/
The order is always: PCH, Infuse, MrMC, Plex.

4 Likes

Very good and interesting post. Thanks for taking all the time necessary for this task.

In my real life viewing I notice that Infuse does an inferior job of upscaling 480p/720p/1080p content to 4K as compared to either the ATV internal player via Plex or MrMC.

Thanks for posting these detailed test results.

Can you post (or send me a PM) with details on which patterns you are using to test color clipping so we can take a further look here? In some cases if the color profile isn’t explicitly defined in the video Infuse will make some assumptions about which to use, so we definitely want to take a look at why the colors in these particular samples aren’t being handled properly.

PM sent.

Do you have any idea why the different decoding results?

Upon further reading with Plex it might be the case that it gets transcoded as it is an mkv file. But this would just answer Plex.

Thanks for sending in your sample. :slight_smile:

Apple makes available a number of different tools and APIs apps can use to play video. This is just one piece of the puzzle as the final picture you see on your screen can be affected based on how apps implement these tools. This means the results from player A may not necessarily match player B or player C - even though all 3 may be using the same Apple provided tools.

1 Like

Gotcha :slight_smile:

Is there any documentation of those tools/APIs publicly available?

Sure.

Check out the Media and Web section at https://developer.apple.com/documentation/

Thanks!

Impressive analisys of the picture quality of those different players. Thanks for helping to make the AppleTV an even better platform

1 Like

I’ve done a Review one the Dune HD Pro 4K II and I used Infuse and my Philips 804 OLED TV to evaluate and compare the picture quality of these devices.

@James maybe you can improve on certain aspects like chroma upsampling or banding in HDR.

I thought I share the comparison here as well:

Chroma upsampling (1080p AVC):

Dune:

Infuse:

Philips:

Neither Infuse nor Philips exhibit banding on the blue bar. Dune however does. It’s unfortunate to see that Realtek still hasn’t got a grip on the banding isse, which I could observe in the Review of the Zappiti 4K HDR Duo with an RTD1295 back in January 2017.

Dune:

Infuse:

Philips:

It looks very similar for all of them. Seems like nearest neighbour for all three players.

Dune:

Infuse:

Philips:

It looks like Infuse and Dune using the simpler nearest neighbour upsampling method whereas the Philips uses one of the superior methods bicubic or bilinear.

Chroma upsampling (2160p SDR HEVC):

Dune:

Infuse:

Philips:

Very hard to see differences.

Dune:

Infuse:

Philips:

Again, only the Dune shows banding in the blue bar. The lower horizontal blue bar shows the chroma upsampling. Again, the Philips uses either Bicubic und Bilinear while Infuse and Dune are using the nearest neighbour.

Dune:

Infuse:

Philips:

The pattern is very fine but its still visible that Philips provides the best chroma upsampling algorithm whereas Infuse and Dune are inferior. Still Infuse is ahead of the Dune in this regard as the lines are a bit smoother.

Chroma upsampling (2160p HDR HEVC, CLL: 600 nits, FALL: 400 nits):

Dune:

Infuse:

Philips:

This time all three of them exhibit banding. However, the Philips is still the best one with very little banding. Infuse is second place and Dune is by far the worst.

Dune:

Infuse:

Philips:

This pattern shows a very surprising and weird result with the Philips and the Dune as there is a yellow shadow around the blue line as well as a green shadow around the red one. Furthermore, the red line becomes pink at button. This should definitely not occur and the picture should look like the previous ones or like Infuse in this example.

Upscaling:

Dune:

Infuse:

Philips:

None of the three players really excels at this pattern as all of them show some double contours and some artefacts. Only the degree to which they show them differs. The least side affects can be seen by Infuse, which does a very decent job on the horizontal and vertical lines. This is the point in which the Realtek SoC is not as god as it tries to sharpen the image too much. This can be seen by the two vertical lines where a white line appears right next to them. The Philips is softer than the Dune, which is a good thing but it still shows double contours as well as some scaling artefacts around the white cross.

Dune:

Infuse:

Philips:

The image of Dune and Infuse is definitely cleaner with less scaling artefacts than the Philips. Dune is the only player which can clearly distinguish the dot of the i. With Infuse and Philips the small i looks almost like a capital I as they try to merge them together.

The “Audio-Video-Technik” writing in general looks much cleaner on the Dune. The same holds true for the “Burosch” Logo.

It’s hard to pick a winner in this comparison. However, the third place clearly goes to the Philips TV.