studiosr 0.1.6

Python library to accelerate super-resolution research

StudioSR

StudioSR is a PyTorch library providing implementations of training and evaluation of super-resolution models. StudioSR aims to offer an identical playground for modern super-resolution models so that researchers can readily compare and analyze a new idea. (inspired by PyTorch-StudioGan)

Installation

From PyPI

pip install studiosr

From source (Editable)

git clone https://github.com/veritross/studiosr.git
cd studiosr
python3 -m pip install -e .

Documentation

Documentation along with a quick start guide can be found in the docs/ directory.

Quick Example

$ python -m studiosr --image image.png --scale 4 --model swinir

from studiosr.models import SwinIR
from studiosr.utils import imread, imwrite

model = SwinIR.from_pretrained(scale=4).eval()
image = imread("image.png")
upscaled = model.inference(image)
imwrite("upscaled.png", upscaled)

Benchmark

• The evaluation metric is PSNR.
• "+" indicates the result of self-ensemble.

Method	Scale	Set5	Set14	BSD100	Urban100	Manga109	Training Dataset
VDSR	x 4	31.860	28.424	27.431	25.729	29.973	DF2K
VDSR+	x 4	31.950	28.491	27.471	25.809	30.182	DF2K
EDSR	x 4	32.452	28.790	27.718	26.635	30.985	DIV2K
EDSR+	x 4	32.612	28.925	27.798	26.859	31.398	DIV2K
RCAN	x 4	32.602	28.825	27.739	26.736	31.127	DIV2K
RCAN+	x 4	32.702	28.940	27.821	27.020	31.563	DIV2K
HAN	x 4	32.567	28.864	27.771	26.767	31.364	DIV2K
HAN+	x 4	32.689	28.940	27.820	26.935	31.687	DIV2K
SwinIR	x 4	32.894	29.066	27.912	27.448	31.947	DF2K
SwinIR+	x 4	32.899	29.117	27.942	27.564	32.147	DF2K
HAT	x 4	32.960	29.206	27.974	27.953	32.409	DF2K
HAT+	x 4	33.075	29.253	28.015	28.087	32.600	DF2K

Method	Scale	Set5	Set14	BSD100	Urban100	Manga109	Training Dataset
VDSR	x 3	34.124	30.155	28.990	27.806	33.109	DF2K
VDSR+	x 3	34.227	30.217	29.029	27.896	33.353	DF2K
EDSR	x 3	34.617	30.510	29.258	28.809	34.116	DIV2K
EDSR+	x 3	34.739	30.652	29.327	29.029	34.470	DIV2K
RCAN	x 3	34.707	30.600	29.297	29.005	34.340	DIV2K
RCAN+	x 3	34.803	30.703	29.362	29.229	34.658	DIV2K
HAN	x 3	34.707	30.610	29.299	29.020	34.368	DIV2K
HAN+	x 3	34.802	30.708	29.367	29.240	34.676	DIV2K
SwinIR	x 3	34.890	30.905	29.457	29.755	35.029	DF2K
SwinIR+	x 3	34.971	30.960	29.479	29.887	35.166	DF2K
HAT	x 3	34.990	31.042	29.522	30.227	35.444	DF2K
HAT+	x 3	35.070	31.092	29.550	30.326	35.571	DF2K

Method	Scale	Set5	Set14	BSD100	Urban100	Manga109	Training Dataset
VDSR	x 2	37.819	33.447	32.102	31.725	38.308	DF2K
VDSR+	x 2	37.891	33.528	32.142	31.836	38.544	DF2K
EDSR	x 2	38.096	33.900	32.341	32.948	39.065	DIV2K
EDSR+	x 2	38.184	34.003	32.387	33.129	39.247	DIV2K
RCAN	x 2	38.167	34.080	32.376	33.160	39.310	DIV2K
RCAN+	x 2	38.222	34.155	32.419	33.388	39.474	DIV2K
HAN	x 2	38.153	34.092	32.370	33.152	39.307	DIV2K
HAN+	x 2	38.210	34.164	32.417	33.383	39.479	DIV2K
SwinIR	x 2	38.292	34.371	32.515	33.788	39.773	DF2K
SwinIR+	x 2	38.366	34.525	32.542	33.936	39.861	DF2K
HAT	x 2	38.471	34.798	32.590	34.401	40.102	DF2K
HAT+	x 2	38.523	34.765	32.624	34.525	40.196	DF2K

License

StudioSR is an open-source library under the MIT license.