A participant at an academic conference stands in front of a research poster titled "LTM: Lightweight Textured Mesh for Efficient Storage and Neural Rendering," co-authored by Jaehoon Choi and affiliated with Meta and the University of Maryland. The poster details innovative methods for efficient storage and neural rendering of 3D meshes. Key sections include an overview of the project, contributions made by the research, and a diagram explaining the pipeline for initializing geometry and appearance fields. The participant, wearing a striped shirt, appears to be closely studying the material, indicative of the subject's technical nature and relevance. The setup includes a laptop on a table nearby, suggesting potential demonstrations or additional data presentations. The setup is marked as poster number 20.
Text transcribed from the image:
20
Meta
University of Maryland
LTM: Lightweight Textured Mesh

Overview
Lightweight Textured Mesh for Efficient Storage and Neural Rendering

Mesh Geometry
Novel-view Synthesis
Goal
Render
Rendering Quality
Scene
Mesh Geometry
Disk size
Storage cost
LTM

Contributions
Utilize classical mesh decimation techniques to extract neural SDFs of large unbounded scenes
A new approach for joint optimization of appearance and geometry 
A trifecta of precise geometry, efficient storage, and comparable rendering quality

Overview of our pipeline

Initialization
Input
Neural Geometry Reconstruction 
Neural Appearance Field
Mesh Generation
Simulation

Initialization of Geometry and

Appearance: Train both appearance a multi-resolution hash positional encoding Geometry: Train the surface geometry a color MLP_following BakedSdf 
Both initializations utilize the contract Appearance MLP learned s