Coursera Learner working on a presentation with Coursera logo and
Coursera Learner working on a presentation with Coursera logo and

As of late, I made a Tensorflow port of pix2pix by Isola et al., shrouded in the article Picture to-Picture Interpretation in Tensorflow. I’ve taken a couple of pre-prepared models and made an intuitive web thing for giving them a shot. Chrome is prescribed.

The pix2pix model works via preparing on sets of pictures, for example, building exterior names to building veneers, and afterward endeavors to produce the relating yield picture from any information picture you give it. The thought is directly from the pix2pix paper, which is a decent perused

Prepared on about 2k stock feline photographs and edges naturally created from those photographs. Creates feline shaded articles, some with bad dream faces. The best one I’ve seen at this point was a feline onlooker.

A portion of the photos look particularly dreadful, I think since it’s simpler to see when a creature looks wrong, particularly around the eyes. The auto-recognized edges are not excellent and by and large didn’t identify the feline’s eyes, aggravating it a piece for preparing the picture interpretation model.

Prepared on a database of building veneers to named constructing exteriors. It doesn’t appear to be secure with how to manage an enormous void region, however, in the event that you put enough windows on there, it frequently has sensible outcomes. Draw “divider” shading square shapes to eradicate things. 

I didn’t have the names of the various pieces of building veneers so I just thought about what they were called.

Prepared on a database of ~50k shoe pictures gathered from Zappos alongside edges produced from those photos naturally. In case you’re great at drawing the edges of shoes, you can attempt to create some new structures. Remember it’s prepared on genuine items, so on the off chance that you can draw more 3D things, it appears to work better.

Like the past one, prepared on a database of ~137k satchel pictures gathered from Amazon and naturally produced edges from those photos. On the off chance that you draw a shoe here rather than a tote, you get a strangely finished shoe.


The models were prepared and sent out with the content from pix2pix-tensorflow. The intelligent demo is made in javascript utilizing the Canvas Programming interface and runs the model utilizing Datasets segment on GitHub. Every one of the ones discharged close by the first pix2pix usage ought to be accessible. The models utilized for the javascript usage are accessible at pix2pix-tensorflow-models.