Interactive Example-Based Terrain Authoring with Conditional

Generative Adversarial Networks

Scientific Paper Presentation

Interactive Example-Based Terrain Authoring with Conditional Generative Adversarial Networks

Interactive Example-Based Terrain Authoring with Conditional Generative Adversarial Networks

Interactive Example-Based Terrain Authoring with Conditional Generative Adversarial Networks

Example session

Interactive Example-Based Terrain Authoring with Conditional Generative Adversarial Networks

• GAN
• deep neural network
• image generation
• trained as a game between generator and discriminator
• cGAN
• uses additional input to direct the generation

Training

Overview

Synthesizers

• Sketch-to-Terrain
• creates terrain from a sketch containing ridges, rivers, altitude cues
• Levelset-to-Terrain
• turns a binary levelset image into a terrain
• Eraser
• removes a user-specified part of the terrain and completes it
• Erosion
• transforms an input terrain into the corresponding eroded terrain

Sketch-to-Terrain

• rivers - simulating water flow
• crest lines - rivers inverted
• altitude cues - points where water accumulated

Sketch-to-Terrain training set

Synthesizers

• Sketch-to-Terrain
• creates a terrain from a sketch containing ridges, rivers, altitude cues
• Levelset-to-Terrain
• turns a binary levelset image into a terrain
• Eraser
• removes a user-specified part of the terrain and completes it
• Erosion
• transforms an input terrain into the corresponding eroded terrain

Levelset-to-Terrain

• thresholding the altitude

Synthesizers

• Sketch-to-Terrain
• creates a terrain from a sketch containing ridges, rivers, altitude cues
• Levelset-to-Terrain
• turns a binary levelset image into a terrain
• Eraser
• removes a user-specified part of the terrain and completes it
• Erosion
• transforms an input terrain into the corresponding eroded terrain

Eraser

• adding a few discs that represent the missing parts

Synthesizers

• Sketch-to-Terrain
• creates a terrain from a sketch containing ridges, rivers, altitude cues
• Levelset-to-Terrain
• turns a binary levelset image into a terrain
• Eraser
• removes a user-specified part of the terrain and completes it
• Erosion
• transforms an input terrain into the corresponding eroded terrain

Erosion

• meta simulation

Amplification

• small scale details can be added to large-scale terrain
• using a super-resolution technique

Interactive authoring response times

Training times

Inexperienced user session

​

Professional artist session

Feedback

• great read
• lots of illustrations

​

• deep learning knowledge required
• content gets repetitive