P
PWM Explorerreference impl
← All principles

Coded Aperture Compressive Temporal Imaging (CACTI) L1-004

Compressive ImagingVideo snapshot compressive sensingδ=3 · standardL_DAG = 1.4

Forward model E

Single-shot compressive video forward model: a temporally varying binary coded mask gates photons per-pixel for each sub-frame, and the detector integrates all B masked sub-frames into a single 2D snapshot exposure.

L-DAG

S.temporal.coded -> int.temporal
S.temporal.codedint.temporal

Well-posedness W

Existence:
true
Uniqueness:
true
Stability:
conditional
κ:
2500

underdetermined B:1 compressive video recovery; binary random temporally-varying masks satisfy RIP-like conditions; sub-pixel mask mismatch is dominant stability risk (10.4x residual ratio observed, EfficientSCI drops 20.58 dB under severe mismatch).

Solvability C

Solver class:
iterative_proximal (GAP-TV, DeSCI) or plug-and-play (PnP-FFDNet) or learned unrolling (EfficientSCI, BIRNAT)
Convergence rate q:
2
Complexity:
O(H * W * B * log(H*W*B)) per iteration for FFT-based proximal steps; O(H*W*B) for denoising priors.

Specs (1)

Spec IDTitleTyped_spec
L2-004CACTI Mismatch-Only Spec (standard SCI video benchmark family)mismatch_only0.45

Treasury (T_k)

Balance: 0 PWM

Minting

Total minted: 0 PWM