VGG19 is a pretrained network that was originally developed to do image classification, but it has gained a lot of popularity for its usefulness in aiding the training of other networks. It is used in style transfers, deep dreaming, image enlargement and numerous other applications. All of these applications extract layers from VGG19 to work. I recently needed to see what was happening in some layers for a style transfer network I am developing. I decided to share the application I built here in hopes it will help others with a similar need. 

Read more

I’m a geek, programmer, scientist, tinkerer, business man, dreamer, father, husband, thinker. I like to experiment, take risks, and see what is possible.

Subpixel Upscaling/Reshuffle Layers are frequently used in image enlargement networks, though their applications extend far beyond that. In the development of v1 of my Neural Enlarge application, I spent a year experimenting with various things seeking better and better results. I had an early model that was impressive that implemented subpixel reshuffle, but it was slow on Tensorflow for Python, and for Tensorflow.js, it was unusable. It crashed every time and I could not find a solution. Therefore, I moved to using standard deconvolution layers for upscaling.

In the development of v2, I decided to revisit subpixel upscaling to see if I could improve the algorithm. Lets take a look at how it works.

Read more

I’m a geek, programmer, scientist, tinkerer, business man, dreamer, father, husband, thinker. I like to experiment, take risks, and see what is possible.