search

w3c / png

Maintenance of the PNG specification
https://w3c.github.io/png/
Other
44 stars 11 forks source link

Liaison letter from ISO TC 42/WG 23 on Gain Maps #366

Open svgeesus opened 10 months ago

svgeesus commented 10 months ago

Liaison_Letter_to_the_W3C_from_ISO_TC_42.pdf

svgeesus commented 10 months ago

(They also sent a draft of the specification, but it has restrictions on public storage. I am asking them to lift that restriction, so we can discuss it)

svgeesus commented 10 months ago

I have been aware of the work since 2021, and have discussed it at a couple of ICC HDR WG meetings; I also saw a demo of this, in Adobe Photoshop and running on an HDR display, at the ICC meeting in London in April this year. However, this is the first time I have seen the actual draft text of the specification, which clears up some questions I had and raises others.

The overall approach is a good one, and it gives excellent results with much less filesize overhead and more flexibility on viewing conditions, compared to storing the two images separately.

The ISO draft spec, in Annex C, has placeholders for gain maps in AVIF, HEIF and JPEG but not for PNG. We should ask that they add one for PNG as well.

svgeesus commented 10 months ago

TC42-reply-20231107.pdf

leo-barnes commented 10 months ago

Not speaking for TC42 here, but I'm heavily involved with the HEIF/AVIF side of things.

The way we're doing it there is that we're adding dedicated signaling to the HEIF spec that describes how the required metadata in 21496-1 shall be stored in a HEIF file. I would then imagine that the HEIF/AVIF sections in 21496-1 would simply refer to the relevant section in HEIF rather than spell out exactly how it works.

Concretely, we propose doing this in HEIF by creating a new derived item type that describes how the inputs (base image and gain map) are combined into a reconstructed image. The metadata needed is stored in the box body of that item using a dedicated structure rather than as XMP which is what the current demo files do.

svgeesus commented 10 months ago

Yes, I would expect 21496-1 to take the same approach with PNG

leo-barnes commented 10 months ago

Not sure how you would be planning on storing this in PNG, but if you decide on taking the approach of having a dedicated structure for the metadata rather than relying on XMP, it might be worth syncing up on it. Would be kind of nice if it's at least roughly the same as what we're proposing in HEIF.

svgeesus commented 9 months ago

For the purposes of discussion, here is a proposal for storing gain maps in PNG.