VESA Display Stream Compression- a new low cost, low latency, near lossless codec

[Procrastinating]

As the title suggests, VESA has developed, and both VESA and MIPI are going to make use of "DSC" for future iterations of Displayport, eDP and DSI. It is intended to be a very low computational cost, low latency, near lossless codec. This is intended to be a solution to the growing bandwidth requirements of higher resolutions, refresh rates and bit depths.

VESA claims that a 24-bit signal can be reduced to "8 bits"(by 66%), and this algorithm can be applied to higher bit-depths as well. At the current displayport bandwidth, this would theoretically enable a "virtual throughput" of 30gbps->90gbps.

DSC is claimed to be "virtually lossless", and the compression should be "unnoticeable" for applications including images/video, "subpixel text"(no chroma subsampling?) and "Problematic patterns".

DSC is fixed-bitrate for obvious reasons.

The whitepaper for DSC can be found here (http://www.vesa.org/wp-content/uploads/2014/04/VESA_DSC-ETP200.pdf)

This could be very interesting if it could also be repurposed to capture and store raw footage in a "near lossless" format, similar to prores or cineform.

[nevcairiel]

It uses YCoCg for color decorrelation, which is at least a neat step in the right direction, and avoids the floating point computations of YCbCr.

I'm not really sure I like the general idea of my image going through a lossy compression before it reaches my screen. But I guess only time will tell how well it works.

[EncodedMango]

Lossy compression at hardware level? That just sounds awful I for one wouldn't want it.

That said it could be a useful codec.

[foxyshadis]

Looks pretty interesting. It's completely lossless on flat and smooth areas that are easy to predict, as well as repeating patterns, and it lowers the bit-depth in more complicated areas where artifacts are less noticeable. Not bad at all, I'd love to take a look at the standard someday.

And they tweaked it in the best possible way: Flipping back and forth asking lots of users to eyeball the differences for dozens of images on dozens of devices. No flawed PSNR-based decisions here.

[EncodedMango]

More compression applied over already compressed content, it doesn't sound too good.

[Asmodian]

This creates an interesting effect in that if you copy the source by recording the output you get a less than perfect copy. Very analoge. ;)

For 4K 10+ bit with current technology cheap chips and cables it makes a lot of sense and if I don't notice and can still copy the source to other devices I might be ok with it.

edit: Does anyone know if it has a generation problem, I assume it does as it is lossy?

Up to 8Kp60 video over DisplayPort, since they started in late 2012 I assume they mean DP 1.2? Isn't that a compression ratio just over 2.2 to 1 (8-bit 7680x4320p60)? It seems plausible to me to be visually lossless and cheap computationally at that low of a compression ratio.

[Rumbah]

Well, as most cinemas use digital projectors nowadays you'll see compressed movies there, too.
And I don't recall seeing artifacts there (it's max 250Mbit/s mjpeg2000). So if it's anything similar I guess that you won't see any difference.

[foxyshadis]

Asmodian, there won't be a generation problem since this is only ever used for the final displayed output. If you ran a background screen capture, it would still capture the real bits, not the output ones, unless you somehow hooked up a way to record the DP stream in-flight. Even in that case I'd expect that you wouldn't get a generation problem because it's based on reconstructed pixels, and once quantized, they should be representable with the same bits in the future. (Aside from any chance of the optimum match changing, depending on how the matching is performed.) According to the whitepaper, the matching is deterministic, so it's not like MPEG where you can get wildly different matches between different encoders and options.

I predict this will look far better than all the crappy 6-bit TN panels that make everything look uniformly terrible (yet movies and games are still watchable even then).

[mandarinka]

You might still encounter it *with* terrible TN LCDs added on top. And many displays with probably keep using 6bit panels with AFRC. In theory, there could be interactions with AFRC dithering, but I guess those could turn up in testing of the compression.