File size: 1,017 Bytes
3c45764
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
---
title: ResShift Super-Resolution
emoji: 🖼️
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.0.0
app_file: app.py
pinned: false
license: mit
---

# ResShift Super-Resolution

Super-resolution using ResShift diffusion model. Upload a low-resolution image to get an enhanced, super-resolved version.

## Features

- 4x super-resolution using diffusion model
- Works in latent space for efficient processing
- Full diffusion sampling loop (15 steps)
- Real-time inference with Gradio interface

## Usage

1. Upload a low-resolution image
2. Click "Super-Resolve" or wait for automatic processing
3. Download the super-resolved output

## Model

The model is trained on DIV2K dataset and uses VQGAN for latent space encoding/decoding.

## Technical Details

- **Architecture**: U-Net with Swin Transformer blocks
- **Latent Space**: 64x64 (encoded from 256x256 pixel space)
- **Diffusion Steps**: 15 timesteps
- **Scale Factor**: 4x

## Citation

If you use this model, please cite the ResShift paper.