This repository provides the Tensorflow implementation for Local Light Field Fusion, a method for novel view synthesis from sparse input images. It is targeted at researchers and practitioners in computer vision and graphics interested in practical view synthesis and light field rendering. The primary benefit is generating high-quality novel views from a limited set of input images.

How It Works

The approach involves converting input images into a local layered representation (MPI) and then blending rendered light fields from these MPIs to synthesize novel views. This method is advantageous as it allows for practical view synthesis with prescriptive sampling guidelines, enabling efficient and high-fidelity results.

Quick Start & Requirements

• TL;DR: Install Docker and nvidia-docker, then run bash download_data.sh followed by sudo nvidia-docker run --rm --volume /:/host --workdir /host$PWD tf_colmap bash demo.sh.
• Prerequisites: Docker, nvidia-docker, CUDA, Tensorflow, COLMAP, ffmpeg. Python packages listed in requirements.txt. Optional: GLFW for OpenGL viewer.
• Setup: Docker build can take 15-30 minutes, or download a ~6GB image.
• Links: Local Light Field Fusion Project (assumed based on description), Paper (assumed based on SIGGRAPH 2019 mention).

Highlighted Details

• Implements novel view synthesis using Local Light Field Fusion.
• Supports generating MPIs from posed images and rendering novel views via CUDA or Tensorflow.
• Includes an interactive OpenGL viewer for exploring rendered scenes.
• Provides guidelines for capturing input images and processing custom datasets.

Maintenance & Community

The project is associated with SIGGRAPH 2019 and lists authors from UC Berkeley and Fyusion Inc. No specific community links (Discord, Slack) or roadmap are mentioned in the README.

Licensing & Compatibility

The README does not explicitly state a license. Given the SIGGRAPH 2019 context and academic authorship, it is likely intended for research use. Commercial use would require careful review of any associated license terms not detailed here.

Limitations & Caveats

The OpenGL viewer is not compatible with the Docker container. Troubleshooting sections mention potential issues with COLMAP and OpenGL initialization on certain macOS versions. The method requires careful camera pose recovery and input image capture for optimal results.