===== Commercial Products =====

==== Atmoph ====
  * There is this [[https://atmoph.com/en|Product that already exists]] which starts at $600ish
    * [[https://www.kickstarter.com/projects/atmoph/atmoph-window-yo-a-smart-virtual-window|Their kickstarter]]
    * a [[https://store.atmoph.com/en/products/sunlight-bar|Sunlight Bar]] is an RGB LED strip that simulates natural sunlight

==== Winscape ====
  * this looks to be [[https://www.youtube.com/watch?v=V2hxaijuZ6w|Exactly the project]], done in 2012!
    * it's worth taking a deep dive into this project. He used 2D video shot from his camera and OpenGL to change the frustrum that's it!

==== Real Life Portal ====
  * [[https://smeenk.com/real-life-portal/|this project]] uses 2 Kinects: 1 outside to capture the depth and 1 inside to capture the person
  * seems very convincing and may be worth a deep dive
  * sometime after 2015 apparently?

==== The Parallax View ====
  * Available on iOS shows nice parallax tracking as a demo
  * by Algomystic AB

==== Lively Wallpaper ====
  * use your mouse to change the view with [[https://apps.microsoft.com/detail/9ntm2qc6qws7?hl=en-US&gl=US|this app]]

==== Looking Glass Studio ====
  * They sell expensive holograms but may have [[https://lookingglassfactory.com/software/looking-glass-studio|free software]] which can handle RGBD files in some sort of 2d "preview" mode

===== Adding movement detection for 3D effect =====

==== Technologies ====

=== Video Data ===
  * Volumetric Video is a video recorded with multiple cameras, allowing 6 degrees of freedom
  * 360 degree "immersive" video is what you see on Youtube. You are standing in only 1 place but can look at various angles.
    * because you cannot move around in space this doesn't seem like it would work.
  * Multi-Angle / Multi-View Video: this is where several cameras capture at the same time. You can jump from camera to camera
    * is it stands this won't work because you are jumping from camera to camera instead of moving smoothly
    * it could be AI would solve this problem somehow but seems beyond the scope, especially in realtime
  * 2.5D video: this is what the roku does in its screensaver. Doesn't seem to fit the bill
  * build simulated video with Nanite using Unreal?

  * [[https://superspl.at/|Splats]] are fantastic but only work for still images
    * based on [[https://www.splats.com/|this site]] maybe this isn't true? There's some reason splats won't work for this...
  * [[https://github.com/DepthAnything/Depth-Anything-V2|Depth Anything]] is open source software that uses AI to change an RGB video to RGBD (D=depth) by creating a depth map
    * [[https://www.owl3d.com/pricing|Owl3D]] is a commercial wrapper for Depth Anything
    * [[https://github.com/nagadomi/nunif]] IW3 is an open source wrapper for Depth Anything
    * with iw3 you'd use ffmpeg to make the side-by-side video. It'd look something like this:
<code>
python run_video.py --video-path garden.mp4 --outdir output --input-size 518
ffmpeg -i garden.mp4 -i garden_depth.mp4 -filter_complex hstack output_rgbd.mp4
</code>
  * [[https://github.com/splatsdotcom/WindowMode|Window Mode]] is open source software that can change your view orientation of a 3D image of video
    * it appears this can change your camera orientation but not the actual camera X,Y,Z position
    * it could be with a landscape view changing only the camera orientation is enough?
  * record video with iPhone LiDAR app (like Record3D)

=== Head Movement ===
  * TrackIR and OpenTrack are both hardware/software combinations that track your head movement via a webcam and translate that movement into the camera angle on your screen
  * Terms to look for: Head-Coupled Perspective, Off-Axis Projection, "Parallax Window" effect
  * As you move, the camera's viewing shape (the frustum) would skew so it looks "through" the monitor at an angle, called **Off-Axis Projection**
  * the famouse [[https://www.youtube.com/watch?v=Jd3-eiid-Uw|Johnny Lee video]]. Skip to around 2'45" for the effect

=== Filling in the missing background ===
This is a combination of "inpainting" (guessing the missing color pixels) with "Depth Completion" (guessing the missing depth values).
  * It's possible to use AI to do this but too slow
  * "LaMa" (Large Mask Inpainting) will be too slow
  * "Mesh Stretching" will be the "good enough" method for landscapes

=== Implementation ===
  * It looks like this would involve a side-by-side video where the left is the video and the right is a depth mapping of the video
  * using shaders and an engine like Unity, Unreal or three.js you'd move the camera around and render the scene

== Realtime ==
  * It looks like it might be possible to take a live video feed, feed it to Depth Anything, pipe that to Unity as a side-by-side with depth mapping on the right
  * concerns: lag, and you need an awesome GPU


== Details ==
in Unity/Unreal/three.js:
  * **Scene**: 3D content (or volumetric video capture) inside a virtual box.
  * **Camera**: placed exactly where your monitor is in the virtual world.
  * **Script**: updates the Camera Projection Matrix every frame based on your head tracking coordinates.

= Unity Example =
  * modify **Camera.projectionMatrix**
  * calculate the corners of your screen relative to your eye position.
  * As your head moves right, the left side of the projection pyramid gets "longer" and the right side gets "shorter," creating the illusion that you are looking through a window.
  * "Vertex Displacement Shader": This is the core logic. "Read the depth map pixel, move the vertex Z position."
  * "Unlit Shader": You don't want the 3D engine's lights to mess up your video lighting.
  * "Shadow Casting Mode: Off": You must turn this off, or your 3D window will cast weird shadows on itself.

===== Easy POCs =====
  * Get a GLB or GLTF file at [[https://sketchfab.com/|Sketchfab]] and load it into Windowmode
  *