When training FSRCNN the output image include some green/red color, Does anyone encounter this?
@Eric suggested that some pixel values may be outside whatever range the display software is assuming for limits. Clipping the outputs to be in valid ranges might fix it. Some additional context as to the model you are running and some code (especially for rendering the image) may help. Thanks!
Thanks @arvoelke I have try clip the output value and even just training the gray image. I still have the problem, looks like some other reason cause this.
Where did you get the FSRCNN model that you are trying to run? Do you see this effect only when using
nengo-dl, and not when using TensorFlow on its own? We will need some details to try to reproduce this (e.g., version of tensorflow, link to model).