---
id: torchsr
version: "1.0.4"
license: MIT
license_treatment: permissive
maintenance: dormant
---

# torchsr — Super Resolution Networks for pytorch

License: permissive · Maintenance: dormant · Downloads: 85.1K/mo

## What it is and what it does

torchsr is a PyTorch library that bundles pretrained models for image super-resolution—the task of increasing image resolution by inferring missing details using trained neural networks. It includes five model architectures (EDSR, CARN, RDN, RCAN, NinaSR) with different size-to-accuracy tradeoffs, ranging from 0.10M to 40.7M parameters. The library also provides common super-resolution datasets (DIV2K, Set5, Set14, B100, Urban100, RealSR, Flicr2K, REDS) with automatic download, data augmentation utilities, and a unified training script.

You load a pretrained model, pass low-resolution images through it, and receive upscaled output. The package is designed for researchers and practitioners who want to apply or benchmark super-resolution without building models from scratch. It depends only on torch and torchvision, making it straightforward to integrate into existing PyTorch workflows.

Use it for:

- Upscale low-resolution photographs or screenshots to higher resolution for archival or display purposes
- Benchmark different super-resolution architectures on standard datasets to compare accuracy and speed tradeoffs
- Fine-tune a pretrained model on custom degraded image data for domain-specific super-resolution tasks
- Integrate super-resolution preprocessing into image analysis pipelines to improve downstream model performance on low-quality inputs
- Research super-resolution techniques by training models on provided datasets with the unified training script

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Provides pretrained neural network models for image super-resolution (upscaling low-resolution images to higher resolution) using PyTorch, including EDSR, CARN, RDN, RCAN, and NinaSR architectures.

Yes, if you need pretrained super-resolution models and can tolerate dormant maintenance. The package is stable, has no known vulnerabilities, and low install friction. However, verify that pretrained weights remain accessible and that it works with your PyTorch version—no updates have landed since August 2022, so compatibility with very recent PyTorch releases is uncertain.

## Install

pip install torchsr
uv add torchsr
poetry add torchsr

## Installing torchsr

Before you install: Low install friction with only torch and torchvision as runtime dependencies. Maintenance is dormant—last release was 2022-08-21 and last commit 2023-12-23, over a year ago, though the repository remains active and unarchived.

License in practice: MIT license is permissive, allowing commercial and private use with minimal restrictions.

Quickstart:

pip install torchsr

from torchsr.models import ninasr_b0
from torchsr.datasets import Div2K

model = ninasr_b0(scale=2, pretrained=True)
dataset = Div2K(root="./data", scale=2, download=True)
hr, lr = dataset[0]

Verify before relying:

- Whether pretrained model weights are still accessible and not hosted on deprecated servers
- Compatibility with recent PyTorch and torchvision versions released after 2023-12-23
- Whether the package will receive maintenance or security updates in the future

## Package facts

- License: MIT (permissive)
- Python support: unspecified
- Install friction: low
- Maintenance: dormant
- Downloads: 85.1K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

image super-resolution pytorch, upscale images neural network, pretrained super-resolution models, EDSR RCAN CARN pytorch, low-resolution to high-resolution, image enhancement deep learning, upsampling neural networks, image-processing, computer-vision, deep-learning

[View on SkillFed](https://skillfed.io/packages/torchsr) · [View on PyPI](https://pypi.org/project/torchsr/)