Expand description
Minimal Inline Deployment (Fast-Path Proxy)
This module implements a reduced per-pixel proxy derived from residual evolution. It is NOT the full DSFB supervisory system. It is a constrained proxy intended to assess deployment feasibility under strict real-time budgets.
The proxy computes per pixel: r_t = L1(C_t - H_t) / 3 residual magnitude d_t = r_t - r_{t-1} drift s_t = d_t - d_{t-1} slew u_t = |d_t| + lambda * |s_t| scalar proxy T_t = saturate(1 - k * u_t) trust
Optionally, u_t is averaged over a 3×3 neighborhood before trust computation.
State: two scalar per-pixel history buffers (residual, drift). Output: one scalar per-pixel trust buffer.
No heap allocation occurs in the inner loop of the CPU reference path.
Structs§
- Fast
Path CpuOutput - Output from one frame of the CPU reference fast-path computation.
- Fast
Path GpuOutput - Result from a single-frame GPU fast-path kernel execution.
- Fast
Path Timing Entry - One row in the fast-path timing study output.
- Fast
Path Timing Study - Full result from
run_fast_path_timing_study.
Constants§
- FAST_
PATH_ K - Default trust slope k. Scales the proxy before saturation.
- FAST_
PATH_ LAMBDA - Default slew weight λ. Applied to |s_t| in the proxy formula.
Functions§
- render_
trust_ strip_ svg - Render a trust field as a simple SVG heat-map strip (one rect per pixel column).
- run_
fast_ path_ cpu - CPU reference implementation of the minimal inline fast-path proxy.
- run_
fast_ path_ timing_ study - Run the GPU timing study for the fast-path proxy at 1080p and 4K.
- try_
run_ fast_ path_ gpu - Execute one frame of the GPU fast-path proxy.