pub fn patchify_video(
x: &[f32],
c: usize,
t: usize,
h: usize,
w: usize,
) -> Vec<f32>Expand description
[C, T, H, W] → [T·(H/2)·(W/2), C·4], the 2×2 spatial patch
flattened channel-major-outer as einsum("nctrhpwq->nthwcrpq").