Hi Theodore - apologies for missing this! A single model is used for all users, which ships as part of Chromium. The generated implementation is checked into the following directory: https://source.chromium.org/chromium/chromium/src/+/main:third_party/blink/renderer/platform/graphics/darkmode/
Thanks, Peter On Mon, Nov 8, 2021 at 11:01 AM Theodore Olsauskas-Warren <sau...@google.com> wrote: > Friendly bump on the above, are the details of the proposed ML model > available anywhere? > > Theo. > > On Friday, October 15, 2021 at 9:57:29 AM UTC+2 Theodore Olsauskas-Warren > wrote: > >> Privacy clarification question, you've mentioned that an ML model will be >> used to distinguish iconography from photos, I'd like to understand a bit >> more about that model. Is the model static? Will we be delivering the same >> model to all users? With what cadence will we be pushing new models? >> >> Thanks, >> >> Theo. >> >> >> >> On Thursday, October 7, 2021 at 6:28:40 PM UTC+2 yoav...@chromium.org >> wrote: >> >>> LGTM to experiment M96-M99 (inclusive) >>> >>> On Thu, Oct 7, 2021 at 5:12 PM Peter Beverloo <pe...@chromium.org> >>> wrote: >>> >>>> Thanks Yoav! >>>> >>>> On Thu, Oct 7, 2021 at 8:15 AM Yoav Weiss <yoav...@chromium.org> wrote: >>>> >>>>> This seems extremely useful! >>>>> >>>>> What are the timelines for experimentation? Do you have partners lined >>>>> up that are ready to play with this? >>>>> >>>> >>>> The experiment will start in M96, and will last until (& including >>>> M99). We have several partners lined up indeed. >>>> >>>> Also, the intent seems to be missing many fields (signals from other >>>>> vendors and developers, TAG review, etc) >>>>> Can you complete the missing ones from the Chrome status entry? >>>>> >>>> >>>> Yikes! Hereby: >>>> >>>> *Blink component* >>>> Blink>DarkMode >>>> >>>> *TAG review* >>>> CSS Color Adjust Level 1 has already seen a TAG review, but >>>> (re)inclusion of the "only" keyword was not part of that: >>>> https://github.com/w3ctag/design-reviews/issues/583 >>>> >>>> We're still pursuing some additions for a more complete developer story >>>> (e.g. https://github.com/w3c/csswg-drafts/issues/6664) and will >>>> request review after that. >>>> >>>> >>>> *TAG review status*Not applicable >>>> >>>> *Interoperability and Compatibility* >>>> >>>> *Gecko*: No signal >>>> *WebKit*: No signal, although Web contents in Apple Mail on iOS >>>> supports a force dark mode. >>>> >>>> >>> Would be good to get such signals, but that's not blocking: >>> https://bit.ly/blink-signals >>> >>> >>>> *Web developers*: Mixed: we've received positive feedback regarding >>>> the decrease in cost of providing a dark mode with the availability of a >>>> per-element opt-out, but have also heard concerns about the quality of the >>>> darkening algorithms. This experiment will allow us to gather more data on >>>> both. >>>> >>>> >>> Makes sense. >>> >>> >>>> >>>> *Is this feature fully tested by web-platform-tests?* >>>> Yes, parsing tests are included in //css/css-color-adjust/. >>>> >>>> *Requires code in //chrome?* >>>> No. >>>> >>>> Thanks, >>>> Peter >>>> >>>> >>>>> >>>>> On Tue, Oct 5, 2021 at 10:18 PM Peter Beverloo <pe...@chromium.org> >>>>> wrote: >>>>> >>>>>> *Contact emails* >>>>>> fut...@chromium.org, p...@chromium.org, pe...@chromium.org >>>>>> >>>>>> *Spec* >>>>>> https://drafts.csswg.org/css-color-adjust/#color-scheme-prop >>>>>> >>>>>> *Summary* >>>>>> This Origin Trial covers two pieces of related functionality: >>>>>> >>>>>> First, websites that decide to participate in this trial will have >>>>>> their appearance algorithmically adjusted to be darkened, and thus >>>>>> appropriate for display whilst a device is in Dark Mode. Rendered colors >>>>>> will have their lightness adjusted to appear dark, and images will be >>>>>> classified, where images will be classified first based on their size and >>>>>> properties, then by a machine learning model to help distinguish >>>>>> iconography from photos and other images. >>>>>> >>>>>> Second, support for the CSS color-scheme “only” keyword, which >>>>>> forbids the user agent from overriding the color scheme for an element. >>>>>> This enables participating websites to manually touch-up certain elements >>>>>> when the algorithm gets it wrong, which will happen. >>>>>> >>>>>> *Goals for experimentation* >>>>>> Modern operating systems, particularly mobile ones, feature >>>>>> extensive, platform-wide Dark Mode implementations. Dark Mode is the >>>>>> default, out-of-the-box experience on some Android devices, and >>>>>> automatically enabled during nighttime on many more. >>>>>> >>>>>> Early data is suggesting that dark themes use 10% less battery than >>>>>> light themes at 75% brightness, and suggest significant differences in >>>>>> user >>>>>> behaviour on websites that don’t provide a dark mode whilst the user’s >>>>>> device is in dark mode. >>>>>> >>>>>> However, we heard feedback from developers that one of the reasons >>>>>> why dark themes haven’t been more widely adopted on the web is the cost >>>>>> of >>>>>> creating and maintaining a second theme for web applications. >>>>>> >>>>>> With this Origin Trial for Auto Dark Mode, we’d like to experiment >>>>>> with both partners and developers to understand the impact of Dark Mode >>>>>> on >>>>>> their websites, and to understand the quality of our algorithms outside >>>>>> of >>>>>> our own testing environment. >>>>>> >>>>>> *Experimental timeline* >>>>>> The experiment will start in M96, and will last until (& including >>>>>> M99). Auto Dark Mode is already programmatically detectable by developers >>>>>> to inspect the computed styles, but we might want to iterate during the >>>>>> Origin Trial with an addition if the CSS working group agrees on one: >>>>>> >>>>>> https://github.com/w3c/csswg-drafts/issues/6664 >>>>>> >>>>>> *Any risks when the experiment finishes?* >>>>>> No, websites will revert back to their original appearance. >>>>>> >>>>>> *Ongoing technical constraints* >>>>>> Auto Dark Mode in particular is a fairly intrusive feature, but >>>>>> builds upon the force dark mode feature already launched for Android >>>>>> WebView. >>>>>> >>>>>> *Debuggability* >>>>>> Developers already have the ability to test this functionality on >>>>>> their pages by opening DevTools, selecting the Rendering option in "More >>>>>> tools", and then emulating Auto Dark Mode. >>>>>> >>>>>> Users and developers alike have the ability to enable this >>>>>> functionality for all websites through a flag which can be found on >>>>>> chrome://flags/#enable-force-dark. >>>>>> >>>>> >>>>> Neat! >>>>> >>>>> >>>>>> >>>>>> *Will this feature be supported on all five Blink platforms supported >>>>>> by Origin Trials (Windows, Mac, Linux, Chrome OS, and Android)?* >>>>>> The Origin Trial will be available for Android only. >>>>>> >>>>>> There are no strict technical reasons for this, but given that the >>>>>> functionality is most appropriate for mobile devices, there might be >>>>>> different considerations between mobile and desktop, and that the team is >>>>>> currently focused on improving the Android experience, we’d like to begin >>>>>> there. >>>>>> >>>>>> *Link to entry on the feature dashboard* >>>>>> https://chromestatus.com/feature/5672533924773888 >>>>>> >>>>>> -- >>>>>> You received this message because you are subscribed to the Google >>>>>> Groups "blink-dev" group. >>>>>> To unsubscribe from this group and stop receiving emails from it, >>>>>> send an email to blink-dev+...@chromium.org. >>>>>> To view this discussion on the web visit >>>>>> https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CALt3x6%3DkTVanrag5ePo39E-FXBzgNChkxqU35CvfQsYZ6AxZeg%40mail.gmail.com >>>>>> <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CALt3x6%3DkTVanrag5ePo39E-FXBzgNChkxqU35CvfQsYZ6AxZeg%40mail.gmail.com?utm_medium=email&utm_source=footer> >>>>>> . >>>>>> >>>>> -- You received this message because you are subscribed to the Google Groups "blink-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+unsubscr...@chromium.org. To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CALt3x6%3Dwg4dTED92Ona47z6nhQey3KGTviBzHxZ4VdL1qzNNaQ%40mail.gmail.com.
