Freedom251 › Image Upscaling Software
BSRGAN
by BSRGAN
What BSRGAN is
BSRGAN is a free PyTorch project for blind image super-resolution and practical degradation modeling. It uses AI upscaling on Linux and Windows, supports batch processing, and accepts JPG, JPEG, PNG, PPM, BMP, and TIF files. The repository notes trained models for 2x and 4x scale factors, with enlargement supported up to 4x. Pretrained BSRGAN and BSRNet model files are available through Google Drive or Tencent Weiyun. Its training approach creates paired low- and high-quality images from high-quality sources. The degradation model can simulate isotropic and anisotropic Gaussian blur, nearest, bilinear, bicubic, and down-up-sampling, plus Gaussian, JPEG-compression, and processed camera-sensor noise. It shuffles degradation order rather than using one fixed sequence. Training instructions cover PyTorch DataParallel and DistributedDataParallel, including a four-GPU example. Settings and degradation types can be changed for specific applications. The project is licensed under Apache-2.0.
Who it suits
BSRGAN suits developers and researchers working with image upscaling, degradation modeling, or custom PyTorch training on Linux or Windows.
What it does well
- Free Apache-2.0 project
- AI upscaling up to 4x
- Batch processing supported
- Multiple image formats accepted
- Pretrained models available
What to know first
- May not suit already degraded low-resolution images
- Training setup can involve distributed PyTorch
- No commercial support or integrations listed
Plans
| Plan | Billed and limits | Price |
|---|---|---|
| BSRGAN | Open-source PyTorch code and pretrained models; no paid plans listed | Free |
As github.com lists them, read 11 Oct 2026. The maker's own page has the final word.
Verdict
BSRGAN combines practical image degradation simulation with AI super-resolution, pretrained models, and configurable training. Its stated limitation for already degraded low-resolution inputs is important when choosing it for restoration work.
Facts from BSRGAN's own pages
- Batch processing
- Yes · github.com, 22 Sept 2026
- Blur
- The model uses isotropic and anisotropic Gaussian blur kernels in both high- and low-resolution spaces. · github.com, 11 Oct 2026
- Customization
- The degradation model can be modified by changing its parameter settings and adding degradation types for a particular application. · github.com, 11 Oct 2026
- Degradation order
- It randomly shuffles degradations instead of always applying blur, downsampling, and noise in a fixed pipeline. · github.com, 11 Oct 2026
- Downsampling
- Supported downsampling methods include nearest, bilinear, bicubic, and down-up-sampling. · github.com, 11 Oct 2026
- Founded
- 2021 · github.com, 28 Sept 2026
- Free plan
- Yes · github.com, 22 Sept 2026
- Input formats
- JPG, JPEG, PNG, PPM, BMP, TIF · github.com, 22 Sept 2026
- License
- The repository identifies its license as Apache-2.0. · github.com, 11 Oct 2026
- Limitation
- The authors say the degradation model tends to be unsuited to model an already degraded low-resolution image because it has many parameters and uses random shuffling. · github.com, 11 Oct 2026
- Maximum scale
- 4x · github.com, 22 Sept 2026
- Noise
- The degradation model includes Gaussian noise, JPEG compression noise, and processed camera sensor noise. · github.com, 11 Oct 2026
- Pretrained models
- The project provides BSRGAN and BSRNet pretrained model files for download through Google Drive or Tencent Weiyun. · github.com, 11 Oct 2026
- Purpose
- BSRGAN is a PyTorch project for blind image super-resolution and practical degradation modeling. · github.com, 11 Oct 2026
- Research origin
- The project presents work titled “Designing a Practical Degradation Model for Deep Blind Image Super-Resolution,” accepted to ICCV 2021. · github.com, 11 Oct 2026
- Scale factors
- The repository notes trained BSRGAN models for scale factors 2 and 4. · github.com, 11 Oct 2026
- Support and integrations
- The page provides code, training instructions, and model download links, and does not list commercial support or product integrations. · github.com, 11 Oct 2026
- Training
- The instructions describe training BSRNet and BSRGAN with PyTorch DataParallel or DistributedDataParallel, including a four-GPU example. · github.com, 11 Oct 2026
- Training data
- Its degradation model synthesizes paired low- and high-quality training images from high-quality images. · github.com, 11 Oct 2026
- Upscaling method
- ai · github.com, 22 Sept 2026
Where BSRGAN stands on Freedom251
- Image Upscaling Software#66 of 99