Neural Upscale 2x

Upscale images by a factor of 2 via deep convolutional neural network in Python: API and pre-trained models, based on the ESPC architecture by Shi et al. 2016 [1].

Two pre-trained models are included: Luminance and RGB, both of which were trained on approximately 33k various images (~6.5Gb). You can, however, use any Tensorflow model which is compatible with the architecture, or even train your own, in which case UpscalingUtilities package and Training.ipynb notebook may be useful.

Requirements: Tensorflow, PIL. Tested on Python 3.10.

---

Quick Guide

1. Import necessary utilities, all dependencies are loaded automatically:

from UpscalingUtilities import *
from Neural_Upscale_2x import *

2. Load desired model. Pre-trained models are either Luminance_Model or RGB_Model. Say you want to use the first one:

model = load_upscaling_model('../Luminance_Model')

3. Load desired image using Tensorflow's function, which returns a PIL image instance:

img = load_img('../img.png')

4. Initialize an upscaler object using the appropriate function. Either luminanceUpscaler() or rgbUpscaler(). In this example:

upscaler = luminanceUpscaler(model)

5. Upscale! Returns an image upscaled by a factor of 2 in both width and height (4 times as many pixels) as a PIL image instance, which you can save as usual:

img_2x = upscaler.upscale(img)
img_2x.save('../title.png')

---

---

References

[1] Wenzhe Shi, Jose Caballero, Ferenc Huszár, Johannes Totz, Andrew P Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1874–1883, 2016.