#visual translation

Pix2Pix Edges2Pichaku Example

New interactive Pix2Pix visual translation example from ml5js transforms your doodles based on a dataset trained on a Pokemon character.

1. Press your mouse to draw a Pikachu on the canvas.

2. Click 'Transfer' button.

3. A colored Pikachu image will appear in ~5s.

4. Click 'Clear' button to clear the canvas and draw again.

You can try it out for yourself here

These two are currently for sale on my website!

https://tassiabianchini.com.br/Collect

Go grab it! 

1.  ‘What do lines do...?’ & ‘No mistake’ Pen on paper, 24 x 32 cm / 9 x 12 in each 2020

Second one from the words people sent me. 

The one who leaves, Tássia Bianchini, Pen on paper, 24 x 31 cm, 2020

Tássia Bianchini, Visual translation - survival, 2018

Uncanny Rd

Project by Anastasis Germanidis and Cristóbal Valenzuela is an online Neural Network semantic painting tool. Trained on visual road data, you can doodle or place preset markers which will be visually translated into another image:

Uncanny Road is an experimental tool for collectively synthesizing a never-ending road using Generative Adversarial Neural Networks. It is based on the pix2pixHD project, published by @nvidia and UC Berkeley (Project, Paper and Code), that allows for photorealistic image-to-image translation. The pix2pix model was trained using adversarial learning on the Cityscapes dataset, containing thousands of street images.

To synthesize street images, draw on the colormap of the scene. Each color represents a different kind of object label (e.g. road, building, vegetation, etc.) that the neural network can understand