Title:
Image-to-Mesh Generation for a Playable 3D Level
Subject:
It is proven that Deep Generative Models are capable of generating meshes based on images. The question we aim to answer is whether these models can also generate playable 3D environments for games. The idea is to create a system that takes a top-down view of an environment in the form of an image and generates a 3D environment in the form of a mesh. We also consider a hybrid design with the help of rule-based geometry generation and the neural approach
for refinement. The goal is to determine if the generated meshes are pliable and coherent.
Challenges:
● Finding the best candidate models for this task.
● Creating synthetic data for the training process, with images as
input and meshes as the ground truth.
● Training and fine-tuning the model.
● Integration in Game Engine and evaluation of playability.
Requirements:
● Good understanding of machine learning and deep learning concepts, familiarity with neural network architectures and generative models.
● Some proficiency in programming with Python (PyTorch is a plus).
● Passion for game design.
● Nice to have: Experience in Unity
