• Content
  • A passthrough for any video?

Hello everyone!
Recently a passthrough mode was released, but it is tricky and not always possible to find right passthrough parameters for most of scenes since the passthrough is color-based. SLRLabs made some research into the video-segmentation and background removal. Some deep-learning techniques may be used to segment out the background leaving only people. For now we work on server implementation and got some testing algorythms for background greenscreeeneng. That does not works perfectly yet, but we want to share some results and hear suggestions for testing on other scenes.

Test videos
http://insights.sexlikereal.com/videos/pass/SLR_fisheye190.mp4
http://insights.sexlikereal.com/videos/pass/SLR_StasyQVR.mp4
Removed background
http://insights.sexlikereal.com/videos/pass/testchunkrecoded_190.mp4
http://insights.sexlikereal.com/videos/pass/testchunkstacy_200.mp4

Any suggestion on passthrough will be appreciated!

    Wow this is impressive! Is the background removal done on the fly?

    How do watch the videos? The first link was the only one video I was able to download 😢

    Man no way! if I can do this on a app on my phone, slr should be able to have the passthrough options work similar to these apps!

    Is this where yall trying to get too with the pt, Getting the pt options to work like these apps? Cuz that would be hella freaking awesome!

    Took me literally like 5 minutes to do this, 5 minutes cause I was watching a video how to do it.

    No way! Smh if yall can get the passthrough options to work like that, omg! 🤯🤯🤯🤯🥂😘

    Edit: tried to play it on deo, it played 👍
    But the video came out low quality, but it works
    Hell, practically took out the dudes body 👍

    But yeah man, I'm sure yall could get the passthrough options to work similar how these apps work. 👍

    I like how it cut the male model out, but I also like to have the male model body there, I like both 🤷‍♂️
    To have options to decide with model or with out would be great too.

      Great job on this one! Research does take time an effort! Keep it up, team!

      Man no freaking way, lol I took a pic of my laying cg position background, put my background, then over laid it with that same video did the "cut out" and sized up the video to fit the pic, bro this how i see passthrough would work on a phone, I forgot who mentioned trying pt for phones but it would work. 👍

      But damn that shit looks crazy, it's like watching a video of a female cg or rcg 1on1 and your bed and background there. Very wild to see. Lol

      But seems very possible to get the deo pt options to work similar to these apps.

      sletje capcut.

      Here's the video tutorial, simple
      https://youtu.be/05eXxFtygx4

      I haven't tried full 8ks, I tried it with trailers.

      But works pretty good just to try it out.

      Aye, when you done, ready to share click on more and send it to ur email drive folder, then download from there then you gotta put it on ur headset, I put it on my computer then to my q2.

      Should be easier way tho.

      Edit: nvm the drive method, after its done exporting it saves to your download folder on your phone.

      Edit:
      You can send it to ur gmail drive and then download the videos from oculus browser by signing into ur Gmail. 😉

        Another one!

        See how it cuts male models chest out, need figure out how not to do that.

        Anyway, the way I can easily passthrough that with a black background u can add green too, i didnt put a bg i just clicked on cut background, would like be able to do that in side slr app with the passthrough options 👍

          There's a big difference between processing an image vs building an AI that can do thousands of videos in real time, all with different LUTs, bitdepths, etc. And work on multiple different platforms.

          There's very good background removal software, but you still need to edit them individually semi-manual.

          The ideal way would be using a depth sensor cam like Azure Kinect or Intel RealSense to record depth maps that then can be used as alpha mask, but there are technical hurdles every step of the way. It also doesn't work retroactively.

          So how can we make an AI that does this with certain accuracy. Teach it all the body parts and movement flow prediction. Remove the rest. - I'm pretty impressed with the computer vision our SLR team has put together so far!

          This AI can then also be used for other things, like censoring of JAV video. I believe this is still mostly done manually now. Generating scripts...

          Vrsumo2017 hopefully that sheds some light on why it's not as simple as it might seem. Instead of identifying background to remove, it needs to identify what not to remove. This is not a one-off tool we're building. This is for making it part of the workflow in SLR, actual development and production.


          Future, this can then be expanded in multi axis toys, hololenses, haptic feedback other than strokers, to eventually something like sex-androids.

          @doublevr has posted each of these things individually all over the place already, but he never bothers to explain it or paint the bigger picture 🙂


          As SLR staff I'm of course biased, but our R&D really sets SLR apart from any other studio!

            Vrsumo2017
            Thanks for the link!
            We are looking for existing solutions too since that will be a shortcut to the greenscreening. Unfortunately the existing background removals or segmentation algorythms are trained on clothed persons, usually not making sex. So they oftenly fail on non-standard for ticktock poses and cloth


            Re-training the algorythms is possible but requires a lot of manually-segmented data into specific domain, so we are trying some creative approaches before going to that pass.
            One of them is depthmap-based selection (depth estimation from stereoimages is domain-agnostic task), but it has some drawbacks (depth treshold selection is not intuitive and some objects are closer than model). Other approach is usage of state-of-art language-image models for natural-language-driven segmentation. Potentially such approach can overcome the limitations of traditional deep-learning segmentation - recent results are promising

            Cool option for the deo pt would be to be able to have my "custom panorama" background pic be layerd behind the vr video I'm watching, I might be able to use my laying cg position background picture and that's how I could probably get color for the passthrough. The bg would be my picture.

            I made a 2d greenscreen video with my laying cg position bg picture and over layerd it with a vr video, I just used one side tho, so zooming and positioning was needed, added sum music, watched it in deo, zoomed in sum and that shit looked crazy, was it perfect? No but it close enough to get a since how it could work. Was cool to make. Literally takes like 5 minutes, plus I was just making short 3 min clips.

            But anyway, when I was watching the video in vr I thought...well maybe if I could use a "custom panorama" and have it behind the vr video as it plays and when i use the deo passthrough options my bg pic is behind it.

            Was just a thought i had how q2 users could maybe get sum bg color.. 🤷‍♂️

            A user commented about this website, it seems more advanced then that capcut but it's a website you have to sign up on to use it, it's free but I don't think I wanna be uploading adult content to their site, but I was thinking if slr had something like this inside slr website the users could make short ar clips and upload them for viewing, that would be fun.

            The program looks easy to use but looks like it works more advanced at making ar.

            It's called runwayml
            https://m.youtube.com/watch?v=5AylJMUVeEs&feature=youtu.be

            I got a question, the deo site, would I be able to use that site as a private user and upload short clips for my own personal use and for my own viewing?

            On that capcut I found out how to keep male models body, instead of choosing "remove background" you choose the chrome key and use the intensity bar.

            See 👀

            That capcut pretty cool and simple to mess with.

            Lol, yall should try the 2d video I did , i mentioned it few comments up...👆, i made a few of em already, when you watch it in vr and zoom in sum it looks like vr but with less fov, if that's what color pass through looks like, wow 👌

            Anyway, if anyone trys to do it hit me up, I give you the details if you need help, its very simple. 👍

            I threw a 8k vid in it and just cut out what i didn't want, so I just made cg rcg 3 min vids, rendered pertty smooth, but over 3 mins gets kinda slower to export.

            👍

            Is this passthrough stuff not possible on a G2? Can't seem to find how to get it to work...

            Vrsumo2017 I like how it cut the male model out

            Hummm 🤔
            So I been making sum passthrough vids and figured out how to leave the male body, idk man kinda starting to like the male models full body in, looks great with the hip thrusting movments and with solid clear penatration. Looks great 👍

            Lol, man I been adding music to my 3 min clips, I add sum trap instrumental beats on low volume tho so it's in the back ground, makes it seem old school porno-ish but in 2022 😁👍

            Got me thinking what if the slr player had options to play copyright free music and volume to adjust. As an option, I'd be into that, I'm sure there's sum directors that be producing music too, could feature their music as well.

            Anyway just a thought o had. 👍

            I might just start making my own 5min clips of my favorite positions and add music in the bg for my own use.

            Edit:
            Just thought something, maybe sum sound effects, lol I added a Lil echo to one of my clips, high as fuck very trippy lol 😆...straight dreaming