My project preforms video convolution that takes in a video to be convoluted and the convolution is based on. The “convolution video” (matrix the convolution is based on) is calculated by taking the average of the RGB values and converting them to a float that is in 0 to 1.
It can be scaled so that the videos can be broken up into multiple pieces that convolve independently of each other. This is done through the poly~ object. In the poly~, I take sub matrixes of each of the videos that correspond to the portion of the video we want to view and convolve them.
In the sample video I provided I convolved the built in basketball video with the built in roller blading video. In my patch I scale the resolution of the video to 80×60. This is because the patch uses math that involves a lot of matrices in real time which is very intensive I spent so much time waiting and having Max crash on me 🙁 . I also included a patch that will scale to any size video but most videos are at a resolution that is too intense so it probably won’t work (but the math checks out). To test the realistic version use super-conv and the project1-main patches.
The sample demonstrates the patch working in 2×2, 2×1, and 1×1 “convolution resolution”.
https://drive.google.com/drive/u/0/folders/1xjNnMdg0R6cHRm8qa02h7HfTatXLNZmL