Immersive Web WG/CG TPAC 2020 virtual Day 1

02 November 2020


alexturn, bajones, cabanier, klausw, Laszlo_Gombos, madlaina-kalunder, mounir, yonet
ada, cabanier, Chris Wilson, cwilso, yonet

Meeting minutes

<ada> Hi

<dom> Quit (Ping timeout: 180 seconds)

<lgombos> Join (~lgombos@ef89345d.public.cloak)

<cwilso> hello!

<cwilso> (running a couple of minutes late)

<ada> sitting a couple of minutes late

<alcooper> Join (~alcooper@ef89345d.public.cloak)

<ada> the problem with not using my fancy camera is that youcan see all my toys

Join (~uid418519@ef89345d.public.cloak)

What is the address of the zoom meeting?

<bialpio> Join (~bialpio@ef89345d.public.cloak)

<madlaina-kalunder> Join (~madlaina-kalunder@ef89345d.public.cloak)

<ada> zoom fell over

<yonet_> Join (~yonet@ef89345d.public.cloak)

<bajones> Join (~bajones@ef89345d.public.cloak)

Incubations and Specs overviews

ada: let's go through incubations and get their status
… where they are and what is happen
… do we need feedback? Is it blocked?
… or is it done and bring into the WG?

<yonet_> Quit (Ping timeout: 180 seconds)

<ada> my internet died again

(Ada's video died)

bajones: I will just start since Ada's video died
… for WebXR, everyone has an idea of the state of things
… so I want to give an update on how it's being used

<yonet> Join (~yonet@ef89345d.public.cloak)

bajones: it has shipped in multiple browsers and there is solid support from framework developers
… lots of sites are using it
… and are making it a core feature. Matterport had a really cool demo where you could walk through a house
… we see small cases. For instance KaiXR whose model is to distribute cardboard to classrooms
… and there is a beatsaber knockoff which has a big following
… there was contest for making interesting sites under 20kb
… this is all really cool and it's good to see that they opt into progressive content so things also work if VR is not available
… most web users don't have a VR set so it also works for them
… newer devices had better performance
… on the chrome side we've seen an uptick in the use of the WebXR since OpenXR came along
… the vast majority is using cardboard from Google's perspective

<ada> is looking forward to these cardboard video experiences to use layers

bajones: we should continue to develop for simple devices and powerful ones
… the pandemic slowed progress on the spec a bit
… the biggest hurdles are still the privacy implications which we're close to resolving
… there isn't much to do to say if it's ready for CR
… it's becoming part of the accepted web. There's work on improving accessibility
… I'm planning on being part of it for the long run

ada: please let us know if you have questions or if there's something that you want to add to the unconference time at the end of the week

<alexturn> Join (~alexturn@ef89345d.public.cloak)

alexturn: I just wanted to echo what Brandon said and it's great that we end up with devices all along the ability chain
… we should add more to the acknowledge sections
… so people can point to what they've done with pride

bajones: yes, that's a great unconference session
… we can work out some systems to add it to the full range of modules

ada: let's talk about anchors
… who are the representatives?

bialpio: I think that would be me. I'm unsure if there's a second person
… there hasn't been much acitivity after the long discussion with Alex and Blair
… mostly about reference space stability
… after that got resolved, we shipped anchors in m85 chrome
… no major issues have been reported after shipping
… the biggest outstanding thing is to move it to the hit test API
… the consensus was to wait to reduce confusion for authors
… I'll look into that soon

ada: are there any blockers?

bialpio: I hope there are none because we have shipped them
… I heard that Edge is looking at implementing them using OpenXR

ada: is there anyone from the edge team on the call? That way there are 2 contacts

alexturn: for hololens, we're going to be implementing the anchor implementation

<klausw> Join (~klausw@ef89345d.public.cloak)

lachlan: I will be happy to become the second author on this spec

bialpio: I heard that bajones wants to bring back cloud anchors

<Lachlan> Join (~Lachlan@ef89345d.public.cloak)

bajones: I don't have a special interest in that but it's a good idea for conversation

ada: most people that hear about anchors think that they are cloud anchors

bialpio: there's an issue about that

<mounir> isn't this excluded by the charter?

<ada> https://github.com/immersive-web/administrivia/issues/126

bialpio: that was issue 126

ada: is this in our charter?

bialpio: I think it's explicitly excludes them

ada: what can we do about that?

<mounir> this could be on the previous version actually

ada: we can always recharter but we have so much on our plate so it's unlikely to be done soon

ada: next item: computer vision

klausw: we have several topics about computer vision. Can we postpone it until then

ada: ok. Let's do image face tracking.
… and depth sensing

<ada> https://github.com/immersive-web/administrivia/tree/main/TPAC-2020

ada: raw image access

bialpio: that is something that I added to the computer vision repo

<alexturn> Quit (Ping timeout: 180 seconds)

bialpio: the prototype of the API is not in the CG

klausw: (garbled)
… I didn't look into face tracking
… I've asked for a home for the image tracking explainer and API
… just a quick overview what it looks like now
… it's similar to what Blair was working on.
… my goal was to make something that works with handheld AR
… but it should work with headsets
… there's an experiment to make it work in chrome canary
… (demoing example)
… is there anyone else who wants to takes this one?
… environment tracking

<bajones> Quit (Ping timeout: 180 seconds)

klausw: we don't have to resolve it now but we want to move forward

<alexturn> Join (~alexturn@ef89345d.public.cloak)

<alexturn_> Join (~alexturn@ef89345d.public.cloak)

klausw: developers are really interested. Right now they use the direct camera access
… this API would be more privacy preserving
… and if there's hardware it will be more efficient

alexturn_: the image tracking we have on hololens is not generic
… we have QR code tracking built-in.
… so maybe we can have a base API and add implementations
… maybe there's image tracking for known classes and we'd be interested on the QR code side

ada: alexturn_ would you be interested in being a contact?

alexturn_: yes

ada: ok. I will set up a new repo for image tracking with Klaus and Alex as the contact s

ada: ok. I will set up a new repo for image tracking with Klaus and Alex as the contacts

klausw: it would be interesting to hear what people want to do with a generic API?

alexturn_: maybe a marker API

ada: real world geometery
… is Trevor here today?

bialpio: I don't think any work was done on this
… the main use case was solved with hit test
… so it was deprioritized but we want to pick it up again

ada: does anyone else want to work on this or be a contact?

alexturn_: OpenXR started discussing this
… we're discussing how to best expose meshes, etc
… especially over the next couple of months we'll be interested in discussing this

<alexturn> Quit (Ping timeout: 180 seconds)

ada: computer vision and raw camera access

bialpio: I believe Blair is listed as the lead
… raw camera access is still in my personal repo
… the only feedback is from Blair and about camera's that give back images at a different rate than the requestanimationframe
… on smartphone AR this is not an issue
… but on headsets this is more complicated

alexturn_: I'm happy to participate here
… for HL 1 and 2, image come in a the media cadence

<bajones> Join (~bajones@ef89345d.public.cloak)

alexturn_: for each frame you get extra data to locate the frame
… the question is how you would align it to the phone model
… in our case we align to the render frame rate
… I'm optimistic that we can come to a resolution

ada: do you have a link to the PR?

<bialpio> https://github.com/immersive-web/computer-vision/pull/4

klausw: I had a comment on the computer vision
… it seems that (garbled)
… should there be more general support for computer vision to set up pipelines

ada: the original thinking was to provide algorithms

klausw: we were going to list the use cases and then design generic analysis pipelines
… this hasn't seen much activity
… but nobody has done this pipeline approach
… if people are happy with individual approach, that's fine but if people want to do the generic one, they should speak up

ada: yes, please raise this as a topic to discuss
… next topic is depth sensing

bialpio: that's me again
… quick update: we have an implementation for a depth sensing API
… it's not publicly available
… it will be done as an origin trial to get feedback from developers
… not many people have looked at it yet
… the only feedback was initially from Rik
… we will learn more on how people interact with the API
… please take a look

ada: what is the difference between depth sensing and real world geometry

bialpio: RWS gives you meshes and planes
… depth sensing is used differently. it can be used for occlusion
… it gives out data that's easier to consume by graphics API
… we could merge them together but I would like to hear what people have to say about it

alexturn_: on our API we have planes and meshes and them people do their hit test against that
… or people use them for occlusion
… when people need all the geometry, they need another tier
… the question is what tiers should exists
… should we talk about the full set of use cases
… or are there other breakdowns. We should have a discussion first

ada: should this be an unconference session?

alexturn_: yes, let's discuss the module breakdown

Ada: next topic, detached elements, but Dean isn't here, so let's come back to that.
… DOM overlays. Klaus?

Klaus: there's an editor's draft out, and an explainer.

<alexturn_> Quit (Ping timeout: 180 seconds)

<bajones> Kudos to Klaus for going out of his way to put together even some minimal slides for the incubations he's covering!

Klaus: there's still some confusion as to how this fits in with XR, so that's on me to provide context.
… It's been in Chrome for handheld AR since v84. Has anyone else shown interest?

<alexturn> Join (~alexturn@ef89345d.public.cloak)

Klaus: a fairly large fraction of the AR sites have started using it.
… (shows demo)
… there are some issues that will not translate well to headsets; that's been clear since the beginning, so I'd really like to get back to that discussion.

<Zakim> ada, you wanted to ask about VR dom overlays

Ada: has there been any additional thought within Google for implementing in VR as well, or just AR?

Klaus: I did an initial prototype in VR, but that's been de-emphasized; technically possible, but not a priority.

Lazlo: we're looking into implementing next year.

Alex: really heartening to see DOM overlay get pickup. Was worried people would do UI in 3D.

Ada: we only have one point of contact for DOM overlays. Any volunteers?

Alex: I'll volunteer.

Lazlo: I can also help out

Klaus: we could have a discussion here about how this would translate to headsets if you like

Alex: +1 to the hotspot topic
… (more interested in headset than VR)

Ada: geo-alignment. Blair or Trevor aren't here.

<alexturn> Quit (Ping timeout: 180 seconds)

Piotr: issue on moving anchor hit-test results.

Piotr: Also, hit tests are more like rays than poses, so need to over-specify what hit test results mean
… right now, the specification doesn't say how to construct poses.

<klausw> FYI, it appears that Blair is at VRST2020 this week: https://twitter.com/blairmacintyre/status/1323296767249682434

Piotr: also, hit-test shipping in Chrome since v81.

Ada: any questions for hit-testing, or other unconference topics?

Ada: on to Layers.

Rik: we're continuing to iterate on the spec, regular calls on this spec. Talking with WebGL WG, got their buyin on our specified behavior. Oculus browser is shipping an experimental version, and getting some feedback (and demos)
… we've recently made some changes, but the spec is basically done in terms of functionality. Still need to flesh out some parts (TAG said explainer is not sufficient). Moved to WG, still waiting for FPWD.
… Edge is interested and working on impl

Klaus: I haven't been following closely, but I think we both have DOM layers...
… but they're not in the Layers spec.

Rik: right. DOM layers have some security challenges, so we started with WebGL layers.

Klaus: would be helpful if you would join the unconf on DOM layers

Ada: added an "intersection of Layers and DOM Overlay" to unconf

Rik: I had a 45-min session on Thursday to discuss

<lgombos> prefers Thursday as well

Ada: on to lighting estimation.

Brandon: poor, poor lighting estimation. Movement on this is slow. We've had a prototype (under a flag) in Chrome for a while. Difficult to gauge support for the API's shape, recently lost Kip as a dedicated contributor. :(
… seems relatively uncontroversial.
… ReflectionMap is biggest API challenge.
… Metal API docs were updated after the Seattle FTF, though, so yay!
… on Chrome side, we're still moving forward.
… difficult to work LE in if you're just doing an additive model. But we'd love to hear opinions from the headset world

Ada: Any questions?\
… Moving on, my faaaaaaaaavorite topic: Navigation!
… Unfortunately, Diego is not on the call. Anyone interested in having an unconference session on Navigation?

Brandon: I don't think there's any movement on this, but perhaps still a good unconf discussion?

Ada: Next topic, performance improvements. Repo assigned to Trevor; is anyone interested in taking this up?

<ada> https://github.com/immersive-web/performance-improvements/issues

Brandon: I think a bunch of this got rolled into Layers; we should revisit the repo and check.

Ada: Sounds good, set an unconf topic for that.

Ada: spatial favicons. Ravi is no longer at ML; Dean is interested, so I'm reassigning to him.

Brandon: [dog barks]

Ada: Web GPU binding

Brandon: everyone is really interested in the WebGPU here; there are a few additons like swap-chain-like primitives.
… That's fed back into the Layers spec, which is to the benefit of both

<ada> : synergy makes me happy

Brandon: hoping to be prototyping in Chrome sometime soon. (Probably only Windows-only to begin with)

Ada. Any other unconference topics?

Brandon: if there are any Layers topics, it's natural to fold WebGPU into them.

<alexturn> Join (~alexturn@ef89345d.public.cloak)

Ada: AR module topic:

Brandon: not much going on here. It was always pretty minimal on top of XR core.
… lots of support from the libraries (3js, Babylon, et al) incorporating this.
… Not much trying to shove AR and VR into the same content, which is good.
… Most developers seem to be choosing appropriately.
… AR is almost used with hit-test; perhaps we should consider merging?
… also frequently merged with DOM overlay. I don't think that should be merged into AR core, since it's mobile-centric, but it's nice to see people taking advantage of that.
… the model-viewer component has been seeing great success. (Doesn't always use WebXR, but gaining momentum.)

<ada> was buying a bath the other day and saw a model viewer AR example, it made me happy

klaus: to follow up on apps choosing appropriately... devs shouldn't assume immersive AR is the same as handset AR.

Ada: sounds like an unconference topic

Alex: on AR module requiring/incorporating hit-test: IO'm happy to say "use of AR module requires hit-test", sounds fine to me. I wouldn't want to require immersive-ar in order to use hit-test.

<Manishearth> is kinda against hit test being required for immersive-ar

Alex: for DOM overlay, I think it's probably critical for AR to get this as broadly as possible. Devs should have no excuse to not do things in another way.

Klaus: that could add a heavy burden to implementation, so I wouldn't want to require this for a new implementation.

<Dave_Hill> Join (~Dave_Hill@ef89345d.public.cloak)

Alex: Would still like it to be a de facto part of the standard.

Ada: Web XR Gamepad

Brandon: not much movement since last year. (And I really miss seeing you all in person!)

<ada> misses seeing everyone in person, does not miss the long flights

Brandon: not really a problem, since we're seeing significant usage and not much negative feedback!
… devs love the representations with the profiles as well
… in 3js, this is like three lines of code. So good.
… openXR is jealous of our repository of models. (Continuing to see new submissions, like the Oculus Touch 3 controllers that ship with the Quest 2)
… this is also another thing we might want to merge back into the core spec.
… API seems to be working well for just about everybody.

<yonet_> Join (~yonet@ef89345d.public.cloak)

Zoom just crashed on me

<cwilso> Can someone pick up note taking until I get reconnected?

<Dave_Hill> Quit (Ping timeout: 180 seconds)

We may need to reformat the blobs for binary glbs
… as the nodes are inconsistent
… it would be handy to add some more meta data because then parts of the implementation can be generated from the metadata

<cwilso> thanks Ada.

brandon: cleaning the models is not a problem

alexturn: just agreeing on the node names would be very handy

<yonet> Quit (Ping timeout: 180 seconds)

Brandon: I tend to agree
… was worried naming the nodes would be an undue burden, but seems that people are willing to do that, and the conventions seem to work.

Rik: we're also looking into providing a hand model for the Hands API.
… we had some issues for the v3 model, because we had animations. We might want to consider adding support for animated buttons.

Ada: on to Hand Input.

Lachlan: Hands is at FPWD. People are using it in the wild; seem to like it, based on quick Twitter survey I did. There are some issues I haven't been through yet, but need to get done.
… will go through later
… API does need TAG review.
… spoke with Rossen (MS rep on TAG), addressed some issues.

<Zakim> ada, you wanted to talk about tag review reqs

<Manishearth> Lachlan: i have some comments on grasp that i hope to post tonight

Alex: we have a session on Thursday

alexturn: Please join us thursday at 1pm PST for Hand input discussion

cabanier: we shipped the hand and the feedback has been good. I am looking forward to the Thursday session. I wish we could ship non-experimental

<bajones> Quit (Ping timeout: 180 seconds)

<Manishearth> is excite for hands d??

ada: the very last thing on the list is privacy and security repo but Travis is not here, is there anyone else working on the repo

brandon: I don't think anyone else is working on the repo. We have been talking to the security group and we had good discussions. It would be nice if we can update the repo

brandon: the arifacts are useful to have in the documentation. It is on us to make sure they are in the repo and implemented off of them.

ada: anyone else?

Ada: this has been the last topic. We have a long list of unconference topics which we will put in the agenda Wednesda, Thursday and Friday
… there is no meetings tomorrow. The unconference topics we will have is acknowledgement, anchors, use cases for image tracking
… depth sensing and module breakdown, dom overlay with hot spots and how it works with HDM
… navigation, which is my favorite, performance improvement, if we need to archive the repo.
… final one is required dependencies. What should be required for different modules
… that is it for this evening. We will talk about XR accessibility documentation on Thursday.
… Does anybody else have anything else?

alexturn: It is great that we have this time for unconference. How would we decide when we slide in this unconference things?

ada: that's a good point. We can do doodle poll.

alexturn: it might be hard to move things around, how can we do something similar to sticky notes exercise.

ada: I am going to spin up a public document now so we can all do this now.

<ada> https://docs.google.com/document/d/1yEzJx7ZioNZne531QmZ8TRZI01kS8sTif_W8PL8j1ls/edit?usp=sharing

ada: this is the first time we have to do this without actually post-its. I've pasted the doc in the irc
… I am going to start writing the headers, feel free to fill out the times that you are interested in.

klausw: unless anyone object objects we can combine the DOM Overlay topics into layers topic.

ada: everyone feel free to change the headers.

<Lachlan> Quit (Ping timeout: 180 seconds)

ada: I think the current description for AR, was it AR use cases?

brandon: I think it was mobile vs headset and controllers vs no controllers.

ada: people mark in the document, which particular sessions they would like to go to, it would be usefull
… I am the only one that is signed for navigation

alexturn: I want to go to all of them but I want to see others who are interested in the topics/

ada: thank you so much for coming everyone, this was super productive, I hope to see you again on Wednesday. Bye.

<ada> shoot we did not invite the bot

<ada> i log everything locally so i'll save out a copy

<alexturn> Quit (Ping timeout: 180 seconds)

<bialpio> Quit (Ping timeout: 180 seconds)

<madlaina-kalunder> Quit (Ping timeout: 180 seconds)

<alcooper> Quit (Ping timeout: 180 seconds)

Quit (Ping timeout: 180 seconds)

Minutes manually created (not a transcript), formatted by scribe.perl version 124 (Wed Oct 28 18:08:33 2020 UTC).