AnyRace

• 

  User's view within the scene

• 

  Some of the images we used for the above scene

• 

  Some of the images we used for the above scene

• 

  Some of the images we used for the above scene

• 

  Bird's eye view of a recreated scene

• 

  Some of the images we used for the above scene

• 

  Some of the images we used for the above scene

• 

  Some of the images we used for the above scene

Race against your friends, anywhere, any time.

Inspiration

Recent advances in computer vision have made 3D scene reconstruction a lot more accessible. With only a few pictures, you can now reconstruct a 3D scene from a few 2D images. There's many applications of this, and one of them is to allow you to race anywhere in the world!

What it does

Users upload images of a scene from different angles. Our project recreates this scene in 3D and puts small, player-driveable cars into the scene. Then, users are scaled down to an ant's size and allowed to drive around in the scene. Other users can join your lobby, allowing you to race against them. First one to the end wins!

How we built it

Go, Three.js, dust3r, websockets

Challenges we ran into

Three.js was unfamiliar for all of us and we spent a lot of time reading documentation. Car-wall collisions were a tricky problem conceptually.

Accomplishments that we're proud of

Driving on the gdc toilets

What we learned

We learned a lot about graphics and computer vision!

What's next for Tabletop cars

We will allow user-defined tracks within the scenes with custom start and endpoints or custom numbers of laps.

Built With

• dust3r
• go
• python
• pytorch
• three.js
• webgl