SlapStack Video Compression — motion as a stream of gate events

Sparse Gabor code, 3060-atom pyramid dictionary, 64×64 grayscale. Amplitudes at the measured knee: 1 sign + 4 magnitude bits per gate. Transmitted per frame: gate births (dictionary address + amp), gate deaths (active index), and 5-bit amp updates — nothing else.

input (64×64)

reconstruction (from quantized stream)

gate events — births / deaths / open

the meter

— kbit/s event stream
— kbit/s raw pixels · —× smaller
PSNR (transmitted vs input)—
gates open—
births / deaths / amp-updates per frame—
gate churn (events ÷ open gates)—
event bits / frame—
intra (full-code) bits / frame would be—
fps—
starting camera…

bits per frame — event stream vs intra full-code vs raw 32,768

Honest limits. Toy Gabor dictionary (not a trained Bet 7 encoder); grayscale 64×64; matching-pursuit births, Landweber amplitude refit. The live falsifier is the churn number: the "motion = sparse gate events" claim is only as good as gate survival between frames. If churn approaches 100%, event coding degenerates to intra coding and the claim dies on this screen. Reconstruction is decoded from exactly the bits the meter counts — quantized amps, nothing hidden. Measured offline (gate_only_reconstruction.py): support carries 10–14.6 dB over amplitude-matched random gates; amplitude knee at 4 magnitude bits; a coarse gate is worth ~5 fine gates.