SCRREAM
收藏资源简介:
SCRREAM数据集由慕尼黑工业大学创建,专注于室内3D场景的密集注释。该数据集包含11个场景,每个场景都通过高分辨率扫描和精确的相机姿态注册,生成详细的3D网格和深度图。数据集的创建过程包括扫描、注册、渲染和映射四个步骤,确保了场景的完整性和准确性。SCRREAM数据集主要应用于室内重建、SLAM、场景编辑和6D姿态估计等领域,旨在为这些任务提供高质量的基准数据。
The SCRREAM dataset, created by the Technical University of Munich, focuses on dense annotation of indoor 3D scenes. This dataset includes 11 scenes, each registered via high-resolution scans and precise camera poses to generate detailed 3D meshes and depth maps. The dataset's creation process consists of four steps: scanning, registration, rendering, and mapping, which ensures the integrity and accuracy of the scenes. The SCRREAM dataset is primarily applied in fields such as indoor reconstruction, SLAM, scene editing, and 6D pose estimation, aiming to provide high-quality benchmark data for these tasks.
SCRREAM 数据集概述
数据集下载链接
-
室内重建与SLAM数据集 & 物体移除与场景编辑数据集:
-
人体重建数据集:
-
姿态估计数据集:
数据集结构
下载后解压至 dataset 文件夹,结构如下:
dataset human_scene01.. human_scene02.. pose_meshes_canonical.. pose_scene01.. pose_scene02.. scene01.. ... scene11.. README.txt
可视化指南
环境要求
使用以下命令安装依赖:
pip install -r requirements.txt
测试环境为 Python 3.9,Windows 10。
室内重建与SLAM数据集 & 物体移除与场景编辑数据集可视化
运行 render_scene.py 脚本,参数为 {dataset_dir} {scene} {traj} {frame},例如:
python render_scene.py dataset scene01 full_00 -1 python render_scene.py dataset scene01 full_00 100
可视化格式为 2x3 图像布局:
| (Ground Truth Depth) | (D435 Depth) | (ToF Depth) | | (RGB with Semantic Mask) | (D435 Error) | (ToF Error) |
人体重建数据集可视化
运行 render_human.py 脚本,参数为 {dataset_dir} {scene} {frame} {view},例如:
python render_human.py dataset human_scene01 0 0 python render_human.py dataset human_scene01 0 -1
可视化格式为 1x3 图像布局:
| (RGB with Semantic Mask) | (RGB with Scanned Human Mesh) | (RGB with SMPL Human Mesh) |
6D姿态估计数据集可视化
运行 render_pose.py 脚本,参数为 {dataset_dir} {scene} {frame},例如:
python render_pose.py dataset pose_scene01 -1 python render_pose.py dataset pose_scene01 100
可视化格式为 1x1 图像布局:
| (RGB with Pose 3D Bounding Box and Mask) |

- 1SCRREAM : SCan, Register, REnder And Map:A Framework for Annotating Accurate and Dense 3D Indoor Scenes with a Benchmark慕尼黑工业大学 · 2024年



