Resulting images with incorrect visual rendering

[GisKamal]

Hello,

First of all congratulations for this impressive work, I decided to use it to improve the resolution of my Pleiades images to 2m resolution, we performed an upsampling operation of some existing images from 0.5m to 2m to build patches for training.

Following an initial test, we trained a model from 360 RGB patches (LR,HR), the result of which on an LR image (the same one used to generate the patches) is relatively improved but shows false colors (close to panchromatic). The joint captures show the input LR image and the result image. How can I correct this problem?

Result:

LR_RGB image :

[remicres]

Hi, to train sr4rs you need a source image (low-resolution) and a target image (high-resolution). I am not sure to understand what you are trying to achieve with your Pleiades images of which the native resolution is already 0.5m.

[GisKamal]

Hello, thanks for your reaction, we used a Pleiades LR image with 2m resolution using only the multispectral bands and a typical Pleiades HR image with 0.5m resolution. Our problem is that the resulting image of the trained model is much less colorized than the HR and LR training images.
The screenshots I've attached show extracts of the result image and the LR image to show you the difference in colorization.

[EmanuelCastanho]

Hi @GisKamal , are you still working on this?