OutCast: Relighting Outdoor Single Images with Cast Shadows

The researchers from Adobe and University College London demonstrated a new method of relighting outdoor images. OutCast uses a single image input and predicts cast shadows in arbitrary novel lighting directions while accounting for shading and global effects such as the sunlight color and clouds.

What's especially impressive about the method is the fact that the network uses only one feature. It collects and joins the input image and several maps, including Normal, Cosine, Shadow, and Direction maps.

You can see the results by moving a slider that represents the sun position:

Check out more examples and how it works here and don't forget to join our new Reddit page, our new Telegram channel, follow us on Instagram and Twitter, where we are sharing breakdowns, the latest news, awesome artworks, and more.