Practical overview
When it fits
Use CycleISP when a research workflow needs camera-pipeline-based synthetic data for RAW or sRGB denoising, or when reproducing one of the repository's documented dataset evaluation paths.
Who it is for
- Computer-vision researchers studying image denoising or low-level image restoration.
- Developers exploring camera-pipeline modeling and synthetic RAW or sRGB training data.
Documented uses
- Generate clean and noisy RAW image pairs from clean sRGB inputs.
- Generate clean and noisy sRGB image pairs by mapping synthetic noisy RAW images through the RAW-to-RGB branch.
- Run documented denoising evaluation workflows on supported DND and SIDD image tasks.
Requirements & limits
- The README specifies an older environment: Ubuntu 16.04, Python 3.7, PyTorch 1.1, CUDA 9.0 and cuDNN 7.5.
- Data generation requires downloaded ISP models and sample images; denoising evaluation requires the relevant task model and dataset images.
- The repository documents evaluation paths for DND RAW and sRGB, and SIDD RAW and sRGB, though the supplied README text ends before the SIDD run commands.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Generate synthetic RAW and sRGB denoising pairs
Prepare sample inputs and use CycleISP's documented scripts to generate paired data for RAW or sRGB denoising.
Before you start
- A CycleISP installation using the README's documented Python 3.7, PyTorch 1.1, CUDA 9.0 and cuDNN 7.5 environment.
- The ISP models placed in ./pretrained_models/isp/.
- Sample MIR-Flickr images placed in ./datasets/some_sample_images/.
Steps
Download the models and sample images from the locations linked in the README and place them in the specified directories.
For RAW denoising pairs, run the command below.
python generate_raw_data.pyFor sRGB denoising pairs, run the command below.
python generate_rgb_data.py
Expected outcome
The RAW workflow generates clean and noisy RAW image pairs; the sRGB workflow generates clean and noisy sRGB pairs.
Guide limitations
- The README describes these as data-generation demos and does not specify further configuration or output-directory details in the documented instructions.
Prepare a documented DND RAW denoising evaluation
Set up the model and DND RAW inputs required by the repository's evaluation script.
Before you start
- A CycleISP installation using the README's documented environment.
- The DND RAW denoising model and DND RAW images.
Steps
Download the DND denoising model and place it in ./pretrained_models/denoising/.
Download the DND RAW images and place them in ./datasets/dnd/dnd_raw/.
Run the command below.
python test_dnd_raw.py --save_images
Expected outcome
The repository's documented DND RAW evaluation script is invoked with image saving enabled.
Guide limitations
- The README provides these setup and run instructions but does not state a specific evaluation result.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- swz30/CycleISP · Official homepage title · retained in its source language