Egocentric scene interpretation
In this paper, we present Egocentric Scene Interpretation, an integrated framework that combines depth and semantic maps for 3D scene geometry, semantics, and appearance reconstruction. Furthermore, we introduce a new metric, called the Egocentric Scene Interpretation (ESI) Score, to evaluate the al...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2023 |
| País: | Colombia |
| Institución: | Universidad de los Andes |
| Repositorio: | Séneca: repositorio Uniandes |
| Idioma: | inglés |
| OAI Identifier: | oai:repositorio.uniandes.edu.co:1992/73444 |
| Acceso en línea: | https://hdl.handle.net/1992/73444 |
| Access Level: | acceso abierto |
| Palabra clave: | Egocentric Understanding 3D Reconstruction Semantic Segmentation Depth Estimation Ingeniería |
| Sumario: | In this paper, we present Egocentric Scene Interpretation, an integrated framework that combines depth and semantic maps for 3D scene geometry, semantics, and appearance reconstruction. Furthermore, we introduce a new metric, called the Egocentric Scene Interpretation (ESI) Score, to evaluate the alignment between geometry and semantics for 3D scene understanding. In addition to the experimental framework, we propose EgoScene, a diffusionbased approach that leverages temporal consistency in feature extraction for egocentric downstream tasks. We compare EgoScene against transformer-based methods. Our method outperforms transformer-based methods by 26% in the ESI score, establishing the state-of-the-art for Egocentric Scene Interpretation. |
|---|