Practical overview
When it fits
Use MIRNet when you need a research implementation for image restoration and can work with its Python/PyTorch environment, datasets and model files.
Who it is for
- Computer-vision researchers exploring image restoration methods.
- Developers seeking a documented PyTorch implementation for denoising experiments or evaluation.
- Practitioners working with image denoising, super-resolution or enhancement tasks.
Documented uses
- Evaluate denoising on SIDD or DND RGB images using the repository’s listed scripts.
- Prepare SIDD-Medium image patches and train the denoising model.
- Use the model’s documented research scope to investigate image super-resolution and enhancement.
Requirements & limits
- The repository states PyTorch 1.1.0 and a tested environment of Ubuntu 16.04, Python 3.7, CUDA 9.0 and cuDNN 7.5.
- Evaluation requires separately downloaded pretrained model files and dataset images placed in repository-specific directories.
- The README’s training note specifies two Nvidia Tesla-V100 GPUs.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Evaluate MIRNet denoising on SIDD
Run the repository’s documented SIDD RGB denoising evaluation and save output images.
Before you start
- The MIRNet repository and its stated Python 3.7, PyTorch 1.1.0 environment, including the listed dependencies.
- A downloaded denoising model placed in ./pretrained_models/denoising/.
- SIDD sRGB images placed in ./datasets/sidd/.
Steps
Set up the environment using the repository’s listed system packages, Conda packages and Python dependencies.
Download the denoising model and place it in ./pretrained_models/denoising/.
Download the SIDD sRGB images and place them in ./datasets/sidd/.
Run python test_sidd_rgb.py --save_images.
Expected outcome
The documented SIDD denoising evaluation runs with image saving enabled.
Guide limitations
- The instructions specify SIDD sRGB evaluation; they do not establish results for other datasets or environments.
- The repository’s stated test environment uses Ubuntu 16.04, Python 3.7, CUDA 9.0 and cuDNN 7.5.
Prepare and train MIRNet for SIDD denoising
Follow the repository’s listed data-preparation and training stages for denoising.
Before you start
- SIDD-Medium training data and SIDD validation images.
- The MIRNet repository, its documented PyTorch environment and the warmup scheduler included in pytorch-gradual-warmup-lr.
- The README notes training with two Nvidia Tesla-V100 GPUs.
Steps
Download the SIDD-Medium dataset.
Generate patches with python generate_patches_SIDD.py --ps 256 --num_patches 300 --num_cores 10.
Place SIDD validation images in ../SIDD_patches/val.
Install the warmup scheduler using the repository’s listed setup.py command.
Run python train_denoising.py.
Expected outcome
The documented training script starts MIRNet denoising training using the prepared SIDD data.
Guide limitations
- The README describes a training setup using two Nvidia Tesla-V100 GPUs and refers users to an issue for changing model parameters.
- The instructions cover SIDD denoising training, not training procedures for every listed restoration task.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- swz30/MIRNet · Official homepage title · retained in its source language