Video Mosaic
A photo mosaic… but with videos. A video comprised of a grid of smaller videos.
The above is one of my first passes at this idea. It certainly works as a proof of concept for the idea, but I’d like to improve on the aesthetics. Beyond the code to get get this working, the look and feel could be tuned by adjusting the dimensions of the video grid and the choice of reference videos. This version looks too “busy” for my liking. I think an improved version would use a simpler master video, like low-res pixel art, which evolves slowly. This would allow for the pixel videos to be larger and more stable, and thus more coherent.
Code is here: https://github.com/dean-weez/video-collage
Basically this is a bunch of numpy matrix manipulation with videos represented as four-dimensional (width, height, channel, time) matrices of integers.
- Create the pixel video index.
- For each video in a pixel video corpus, calculate the average over width, height, and time dimensions, yielding a single (r, g, b) value summarizing the video.
- Rescale the master image to match the dimension of video pixels in the output.
- For each frame t of the master video:
- For each pixel:
- Find the index value in the pixel video corpus that is closest to the pixel’s rgb value.
- Select the frame t from this pixel video and set it into the position of the master video’s pixel.
- For each pixel: