Fake 3D
Fake 3D visualization allows creating a pseudo-3D effect from a regular image. Using Parallax Mapping, we implemented this not only for images but also for videos, using a special file format that includes a depth channel in addition to the color channel.
To generate videos in this format, we developed an application that analyzes image depth using Open Neural Network Exchange models.
In addition to classic video parameters such as final resolution, FPS, CRF (Constant Rate Factor) for ffmpeg, and total duration, special attention is paid to temporal stability settings — specifically, the temporal batch size and the number of overlapping frames, which determines how many frames from one batch are considered in the next.
Many of you have seen in movies, music videos, and other clips the technique of compositing augmented reality elements as a secondary layer over regular video.
In our slider application, we implemented the ability to add and edit elements of such a layer: two types of boxes, planes, attach static and animated SVG illustrations (Lottie and Rive players supported), and fully configure text mesh parameters.
As a result, we get a photographic-quality interactive scene with a fully interactive secondary element layer and a pseudo-3D interactive layer for the video itself.
By the way, this was our first experience developing complex real-time interactive visualization based purely on verbal ideas and static references.