Skip to content

Updated attributes to RawNIND model to make raw denoise feature more universal #21

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
TurboGit merged 3 commits into from
Apr 27, 2026
Merged

Updated attributes to RawNIND model to make raw denoise feature more universal #21

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f935290
Update RawNIND model attributes
andriiryzhkov Apr 22, 2026
c1fa93f
Add RawNIND-spesific attributes
andriiryzhkov Apr 23, 2026
923fec3
Switch to preupsample variant and align preprocessing to upstream
andriiryzhkov Apr 27, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 22 additions & 7 deletions models/rawdenoise-nind/model.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,21 +14,36 @@ attributes:
variants:
bayer:
onnx: model_bayer.onnx
input_kind: packed_bayer # 4ch R/G1/G2/B, half-resolution, pre-demosaic
# bayer_v1 contract — see src/common/ai/restore.h.
# policy knobs match upstream RawNIND inference
input_kind: bayer_v1
bayer_orientation: force_rggb # ch0 = R on any CFA
edge_pad: mirror_cropped # reflect in the RGGB-forced frame
wb_norm: none # do NOT WB before inference
output_scale: match_gain # scalar mean ratio
linear:
onnx: model_linear.onnx
input_kind: lin_rec2020 # already demosaicked (X-Trans, Foveon, etc.)
# linear_v1 contract — see src/common/ai/restore.h.
# 3ch full-res lin_rec2020 (X-Trans, Foveon, external demosaic)
input_kind: linear_v1
input_colorspace: lin_rec2020
wb_norm: none # do NOT WB before inference
output_scale: match_gain # scalar mean ratio
target_mean: null # disable exposure boost

repo:
submodule: vendor/rawnind_jddc

checkpoints:
# DenoiserTrainingBayerToProfiledRGB_4ch_2024-02-21-bayer_ms-ssim_mgout_notrans_valeither_-4
# canonical "Bayer" base variant (in_channels=4, funit=32, linear Rec.2020 target)
- url: "https://drive.google.com/file/d/1dFTLeljWi9wwojcZUsam8bE31JdYy3oM/view?usp=drive_link"
# DenoiserTrainingBayerToProfiledRGB_4ch_2024-03-11-bayer_ms-ssim_mgout_preupsample_notrans_valeither_-2
# preupsample variant — needs `preupsample: true` in convert step
- url: "https://drive.google.com/file/d/1xgHafgqRC8ZEGbbXtduSlx9WunD0mgS_/view?usp=drive_link"
path: "temp/rawdenoise-nind/denoiser_bayer2prgb_utnet2.pt"
# DenoiserTrainingBayerToProfiledRGB_4ch_2024-02-21-bayer_ms-ssim_mgout_notrans_valeither_-4
# half-res alternative — ~4× faster, marginally lower quality
# - url: "https://drive.google.com/file/d/1dFTLeljWi9wwojcZUsam8bE31JdYy3oM/view?usp=drive_link"
# path: "temp/rawdenoise-nind/denoiser_bayer2prgb_utnet2.pt"
# DenoiserTrainingProfiledRGBToProfiledRGB_3ch_2024-10-09-prgb_ms-ssim_mgout_notrans_valeither_-1
# canonical "Linear RGB" base variant (in_channels=3, funit=32, lin-Rec.2020 in/out)
- url: "https://drive.google.com/file/d/1kH8tK4RN_edak3r_VIAhCzPa6CNsuTYC/view?usp=drive_link"
path: "temp/rawdenoise-nind/denoiser_prgb2prgb_utnet2.pt"

Expand All @@ -51,7 +66,7 @@ convert:
in_channels: 4
funit: 32
activation: "LeakyReLU"
opset: 17
preupsample: true
fp16: false
- script: convert.py
args:
Expand Down
Loading