denoiser bench — runtimes (auto, WGSL, WebNN, kernels, ORT)
Cold and warm timings, per-stage stats and fp16-vs-fp32 PSNR on deterministic noise images. Pick a runtime and precision, then run every scenario.
Run
| scenario | tiles | build ms | cold ms | warm ms (median) |
upload | encode | run | resolve | PSNR vs fp32 |
noisy input (last)
denoised (last)
results JSON