State-of-the-art image super resolution models for PyTorch.
Project description
super-image
State-of-the-art image super resolution models for PyTorch.
Requirements
super-image requires Python 3.6 or above.
Installation
With pip
:
pip install super-image
Quick Start
Quickly utilise pre-trained models for upscaling your images 2x, 3x and 4x. See the full list of models below.
from super_image import EdsrModel, ImageLoader
from PIL import Image
import requests
url = 'https://paperswithcode.com/media/datasets/Set5-0000002728-07a9793f_zA3bDjj.jpg'
image = Image.open(requests.get(url, stream=True).raw)
model = EdsrModel.from_pretrained('eugenesiow/edsr-base', scale=2)
inputs = ImageLoader.load_image(image)
preds = model(inputs)
ImageLoader.save_image(preds, './scaled_2x.png')
ImageLoader.save_compare(inputs, preds, './scaled_2x_compare.png')
Pre-trained Models
Pre-trained models are available at various scales and hosted at the awesome huggingface_hub
. By default the models were pretrained on DIV2K, a dataset of 800 high-quality (2K resolution) images for training, augmented to 4000 images and uses a dev set of 100 validation images (images numbered 801 to 900).
The leaderboard below shows the PSNR / SSIM metrics for each model at various scales on various test sets (Set5, Set14, BSD100, Urban100). The higher the better.
Scale x2
Rank | Model | Params | Set5 | Set14 | BSD100 | Urban100 |
---|---|---|---|---|---|---|
1 | msrn-bam | 5.9m | 38.02/0.9607 | 33.63/0.9177 | 32.20/0.8998 | 32.08/0.9276 |
2 | edsr-base | 1.5m | 38.02/0.9607 | 33.57/0.9172 | 32.21/0.8999 | 32.04/0.9276 |
Scale x3
Rank | Model | Params | Set5 | Set14 | BSD100 | Urban100 |
---|---|---|---|---|---|---|
1 | msrn-bam | 5.9m | 35.16/0.9410 | 30.97/0.8574 | 29.67/0.8209 | 29.31/0.8737 |
2 | edsr-base | 1.5m | 35.04/0.9403 | 30.93/0.8567 | 29.65/0.8204 | 29.23/0.8723 |
Scale x4
Rank | Model | Params | Set5 | Set14 | BSD100 | Urban100 |
---|---|---|---|---|---|---|
1 | msrn-bam | 5.9m | 32.26/0.8955 | 28.66/0.7829 | 27.61/0.7369 | 26.10/0.7857 |
2 | edsr-base | 1.5m | 32.12/0.8947 | 28.60/0.7815 | 27.61/0.7363 | 26.02/0.7832 |
3 | a2n | 1.0m | 32.07/0.8933 | 28.56/0.7801 | 27.54/0.7342 | 25.89/0.7787 |
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for super_image-0.1.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 0585616646033282cdfdfa3ac60465bb45af68d4c6f807c9d7ed38b1c96dd208 |
|
MD5 | d079f02b1d3977d86e3970c8ff39d8a4 |
|
BLAKE2b-256 | a274769aae387d300e0b4a2563ebb431f13ceb66e7ab87208de9c1effdae81b1 |