Beacon3D
收藏资源简介:
Beacon3D是一个新颖的3D视觉语言(3D-VL)模型基准测试和评估协议,涵盖了3D grounding和question answering(QA)任务,具有对象中心评估框架和用于研究任务一致性的链分析。
Beacon3D is a novel benchmark and evaluation protocol for 3D vision-language (3D-VL) models. It covers 3D grounding and question answering (QA) tasks, and features an object-centric evaluation framework alongside chain analysis for investigating task consistency.
Beacon3D数据集概述
数据集简介
- 名称:Beacon3D
- 类型:3D视觉-语言理解基准测试
- 主要任务:3D grounding(基础定位)和question answering(问答)
- 特点:
- 对象中心评估框架
- 任务连贯性的链式分析
数据集内容
- 测试数据:位于
data/{domain}目录下,包含以下领域:- scannet
- 3rscan(即将发布)
- multiscan(即将发布)
- 元数据:记录每个对象的grounding chains和grounding-QA chains
数据格式处理
- 转换脚本:
grounding_to_scanrefer_format.py:将元数据转换为ScanRefer格式(用于grounding)qa_to_scanqa_format.py:将元数据转换为ScanQA格式(用于QA)
- 输出文件:转换后的json文件(不带
metadata前缀)
评估方法
- 评估脚本:
evaluate_grounding.py:评估grounding任务evaluate_qa.py:评估QA任务
- 评估前准备:
- 实现
extract_pred函数处理模型原始推理结果 - 设置OpenAI API密钥(用于QA评估)
- 实现
- 链式分析:需提供处理后的grounding结果路径
排行榜
ScanNet: QA
| 模型 | Class | App. | Geo. | Spa. | Exi. | Overall (Case) | Overall (Obj.) |
|---|---|---|---|---|---|---|---|
| SceneVerse | 26.4 | 40.4 | 40.0 | 35.0 | 54.1 | 40.5 | 4.7 |
| LEO | 16.4 | 39.8 | 47.6 | 52.8 | 54.3 | 45.2 | 7.5 |
ScanNet: Grounding
| 模型 | Class | App. | Geo. | Spa. | Overall (Case) | Overall (Obj.) |
|---|---|---|---|---|---|---|
| SceneVerse | 73.4 | 65.3 | 61.6 | 73.0 | 73.4 | 51.4 |
即将发布
- 3RScan
- MultiScan
引用
bibtex @inproceedings{huang2025unveiling, title={Unveiling the Mist over 3D Vision-Language Understanding: Object-centric Evaluation with Chain-of-Analysis}, author={Huang, Jiangyong and Jia, Baoxiong and Wang, Yan and Zhu, Ziyu and Linghu, Xiongkun and Li, Qing and Zhu, Song-Chun and Huang, Siyuan}, booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, year={2025} }




