Remove Space-side median from normal output (kept for depth) — user wants raw normals, client-side filter if needed

Rebuild: fp16 + scipy.ndimage.median_filter(ksize=15/11) — staircase fix + FLUX 16-px grid removal

Rebuild: fp16 + scipy.ndimage.median_filter(ksize=15/11) — staircase fix + FLUX 16-px grid removal

Use scipy.ndimage.median_filter ksize=15 (cv2 float capped at 5; grid needed kernel spanning the 16px period)

Use scipy.ndimage.median_filter ksize=15 (cv2 float capped at 5; grid needed kernel spanning the 16px period)

FP16 + median: re-enable degrid_median(ksize=5) to also attenuate the FLUX 16px patchify grid (fp16 already fixed the bf16 staircase)

FP16 experiment: float16 transformer, median degrid disabled to expose raw precision

FP16 experiment: float16 transformer, median degrid disabled to expose raw precision

De-grid: replace ineffective FFT notch with per-channel float median (72%% grain reduction, corr 0.9998)

De-grid: notch-filter FLUX-VAE 8px patch grid from prediction (validated, ~no detail loss)

Combined depth+normal: load both task adapters once, run both passes per call, export raw .npy + 16-bit PNG

Combined depth+normal: load both task adapters once, run both passes per call, export raw .npy + 16-bit PNG

Combined depth+normal: load both task adapters once, run both passes per call, export raw .npy + 16-bit PNG

Export raw float32 .npy + 16-bit PNG as downloadable outputs (full-precision geometry)

Update requirements.txt

Update README.md

Update requirements.txt

Update README.md

Update app.py

Update app.py

Update app.py

minor

init

minor

Update app.py

add login

init

Upload 17 files

initial commit