nind-denoise

A pytorch based image denoising tool for the removal of noise from real photographs. Implements the models developed in
Benoit Brummer's NIND Denoise, and the Darktable workflow pioneered by Huy Hoang.

This fork makes their work easier to experiment with and accessible to a wider audience. Notable features include:

• One-click setup: Download the Lua script and let Darktable handle all Python dependencies automatically
• Support for most hardware via torch.accelerator
  • CPU-only (universal), nVidia GPU (CUDA), Intel XPU/GPU, and AMD GPU acceleration
  • Intel Xe graphics are about 6x faster than CPU, though slower than CUDA
• Automatic darktable integration with export workflow
• Automatic model download and environment setup
• Processed images automatically imported and grouped with originals in Darktable library

Quick Start (Recommended)

The easiest way to use nind-denoise is through the Darktable Lua plugin with automatic environment setup.

1. Install Lua Scripts and Download This Script

First, install the community lua-scripts collection through Darktable's built-in interface (this will create the necessary folder structure). See the official lua-scripts repository for installation instructions.

Once installed, download nind_denoise_rl.lua from this repository and place it in the contrib folder:

Linux:

curl -o ~/.config/darktable/lua/contrib/nind_denoise_rl.lua \
  https://raw.githubusercontent.com/commreteris/nind-denoise/lua/src/lua-scripts/nind_denoise_rl.lua

macOS:

curl -o ~/Library/Application\ Support/darktable/lua/contrib/nind_denoise_rl.lua \
  https://raw.githubusercontent.com/commreteris/nind-denoise/lua/src/lua-scripts/nind_denoise_rl.lua

Windows:

Invoke-WebRequest -Uri "https://raw.githubusercontent.com/commreteris/nind-denoise/lua/src/lua-scripts/nind_denoise_rl.lua" `
  -OutFile "$env:LOCALAPPDATA\darktable\lua\contrib\nind_denoise_rl.lua"

Or manually download the file and copy it to the contrib folder:

• Linux: ~/.config/darktable/lua/contrib/
• macOS: ~/Library/Application Support/darktable/lua/contrib/
• Windows: %LOCALAPPDATA%\darktable\lua\contrib\

2. Set Up Python Environment (One-Time Setup)

1. Start Darktable and enable the script from the Script Manager (lighttable > script manager)
2. In the lighttable module, find the "Update Environment" button
3. Click it and wait while the script automatically:
   • Installs the uv package manager (if needed)
   • Creates a Python virtual environment
   • Downloads the denoising model (~50MB)
   • Installs all required dependencies (PyTorch, GMic, tifffile, etc.)

Note: First-time setup can take several minutes depending on your internet connection and system. The script will display status updates. This is a one-time process.

3. Start Denoising

1. Select one or more images in the lighttable
2. Go to the export module
3. Choose "NIND-denoise RL" as the target storage
4. Configure export settings:
   • Select JPEG or TIFF output format
   • Adjust RL deblur parameters (sigma, iterations)
   • Check "import to darktable" to automatically import processed images back to your library
5. Click export

Processed images will be automatically grouped with their originals in your Darktable library!

---

Advanced Usage

Command-Line Interface

For advanced users, custom workflows, or batch processing, you can use the command-line interface directly.

Basic Usage

To denoise an image, run:

$ python3 src/denoise.py "/path/to/photo0123.RAW"

Note: On Windows, if you use forward slashes do not use single forward slashes for paths. Double backslashes are OK:

PS> python3 src\\denoise.py "\\good\\path\\to\\photo0123.RAW"

Full Command-Line Options

"""
Usage:
    denoise.py [-o <outpath> | --output-path=<outpath>] [-e <e> | --extension=<e>]
                    [-d <darktable> | --dt=<darktable>] [-g <gmic> | --gmic=<gmic>] [ -q <q> | --quality=<q>]
                    [--nightmode ] [ --no_deblur ] [ --debug ] [ --sigma=<sigma> ] [ --iterations=<iter> ]
                    [-v | --verbose] <raw_image>
    denoise.py (help | -h | --help)
    denoise.py --version

Options:
  -o <outpath> --output-path=<outpath>  Where to save the result (defaults to current directory).
  -e <e> --extension=<e>                Output file extension [default: jpg].
  --dt=<darktable>                      Path to darktable-cli. Use this only if not automatically found.
  -g <gmic> --gmic=<gmic>               Path to gmic. Use this only if not automatically found.
  -q <q> --quality=<q>                  JPEG compression quality [default: 90].
  --nightmode                           Use for very dark images. Normalizes brightness before denoise [default: False].
  --no_deblur                           Do not perform RL-deblur [default: false].
  --debug                               Keep intermediate files.
  --sigma=<sigma>                       Sigma to use for RL-deblur [default: 1].
  --iterations=<iter>                   Number of iterations for RL-deblur [default: 10].
  -v --verbose
  --version                             Show version.
  -h --help                             Show this screen.
"""

Manual Python Environment Setup

If you prefer to set up the Python environment manually (not using the Darktable auto-setup), follow these steps:

1. Clone this Repository

git clone https://github.com/commreteris/nind-denoise.git
cd nind-denoise

2. Create Virtual Environment

The variant-enabled version of uv automatically installs the correct version of PyTorch for your GPU.

Windows:

PS> powershell -c { $env:INSTALLER_DOWNLOAD_URL = 'https://wheelnext.astral.sh'; irm https://astral.sh/uv/install.ps1 | iex }
PS> uv venv
PS> .venv/Scripts/activate
(nind-denoise) PS> Get-Command python

CommandType   Name          Version    Definition
-----------   ----          -------    ----------
Application   python.exe    3.1x.xx    C:\Users\<user>\...\nind-denoise\.venv/scripts\python.exe

Make sure python.exe is inside the .venv directory before proceeding.

Linux/macOS:

curl -LsSf https://astral.sh/uv/install.sh | INSTALLER_DOWNLOAD_URL=https://wheelnext.astral.sh sh
uv venv
source .venv/bin/activate
which python
# Should show: /path/to/nind-denoise/.venv/bin/python

Make sure the Python path is inside the .venv directory before proceeding.

3. Install Python Packages

$ uv pip install -r requirements.in --upgrade

4. Verify GPU Acceleration (Optional)

(nind-denoise) $ python
>>> import torch
>>> torch.accelerator.is_available()
True
>>> torch.cuda.is_available()  # For nVidia GPUs
True
>>> torch.xpu.is_available()   # For Intel GPUs
False

---

Requirements

For Quick Start (Darktable Integration)

• Darktable installed and working
• Internet connection for first-time setup
• That's it! The Lua script handles everything else automatically.

For Manual/Advanced Setup

Software

• Darktable and raw images processed with darktable (with .xmp sidecar files)
• GMic CLI (download here for Windows)
• Variant-enabled uv package manager
• Proper GPU drivers with OpenCL support

Darktable Version

• If this PR is still open, you may need to use version 5.2.1 for RL-deblur functionality.

GPU Drivers

nVidia:

• Driver + CUDA toolkit

AMD:

• Driver + ROCm

Intel:

• Intel GPU drivers are slightly trickier due to version mismatches. Install minimum system packages and let uv/pip handle dependencies in the venv.
• See Intel's driver guide
• Install libze-dev and intel-ocloc for PyTorch support
• On Arch Linux: intel-compute-runtime, level-zero-loader, level-zero-headers
• Minimum Linux kernel: 6.14

Verify GPU setup:

$ clinfo | grep device
$ darktable-cltest

---

Citation

Please cite Benoit Brummer's original work:

@InProceedings{Brummer_2019_CVPR_Workshops,
author = {Brummer, Benoit and De Vleeschouwer, Christophe},
title = {Natural Image Noise Dataset},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}