GenMOP

为了验证所提出方法的有效性，我们收集了一个称为通用无模型对象姿势数据集 (GenMOP) 的数据集。GenMOP数据集由10个对象组成，从像 “剪刀” 这样的扁平对象到像 “椅子” 这样的薄结构对象，如图7所示。对于每个对象，同一对象的两个视频序列被收集在不同的环境中，例如背景和照明条件。每个视频序列被分成200个图像。对于每个序列，我们应用COLMAP [49] 分别重建每个序列中的相机姿势，并在对象上手动标记关键点以进行跨序列对齐。To verify the effectiveness of the proposed method, we collected a dataset named the Generalized Model-Free Object Pose Dataset (GenMOP). The GenMOP dataset comprises 10 objects, spanning from flat objects such as "scissors" to thin structured objects such as "chairs", as illustrated in Figure 7. For each object, two video sequences of the same object were captured under distinct environments, including varying background and lighting conditions. Each video sequence is divided into 200 images. For each sequence, we employed COLMAP [49] to separately reconstruct the camera poses in the sequence, and manually annotated keypoints on the objects to facilitate cross-sequence alignment.

搜集汇总

背景与挑战

背景概述

GenMOP是一个用于验证方法有效性的通用无模型对象姿势数据集，包含10个对象，每个对象在不同环境下采集两个视频序列，每个序列含200张图像，并利用COLMAP重建相机姿势和手动标记关键点。该数据集由香港大学、腾讯、德州农工大学和浙江大学于2022年发布，相关论文可在arXiv获取。