https://arxiv.org/pdf/2211.15654.pdf

https://arxiv.org/pdf/2211.15654.pdf

1. Introduction

pre-trained text,image embedding models → 3D-scene Understanding

1. Related work

• Closed-set 3D Scene Understanding.
• Open-Vocabulary 2D Scene Understanding
• Zero-shot Learning for 3D Point Clouds.

1. Methodology

Overview

3.1. Image Feature Fusion

• Image Feature Extraction

Segmentation model:

LSeg, OpenSeg

Input: RGB Image with resolution H x W