Objective

As the mentor for this project, I provided guidance in developing a system for creating editable 3D models of real-world objects using a minimal set of images, with a focus on leveraging Structure from Motion and Neural Radiance Fields.

Technology Stack

• NeRF
• Tensorflow
• Pytorch
• Deep Learning

Key Features

The project aims to revolutionize 3D reconstruction in virtual space, finding applications in scene rendering and self-driving cars. Key features include:

• User-Input Image Processing: Users provide images of the object, initiating the process.
• Structure from Motion (SfM): Determines the camera’s relative position from the object, calculating relative poses.
• Neural Radiance Fields (NeRF): Utilizes deep learning concepts to output RGB color and opacity alpha of each voxel in 3D space.
• Rendering Equation: Integrates NeRF output to render the item’s volume in the 3D space.
• Editable 3D Models: Creates editable 3D models from limited input images.

This project, powered by NeRF, Tensorflow, Pytorch, and Deep Learning, stands at the forefront of 3D reconstruction technology, offering a versatile solution for generating detailed and editable 3D models from minimal visual input.