FewSOL

FewSOL数据集是为机器人环境中的少样本物体学习而设计的，包含336个真实世界物体，每个物体有9张RGBD图像，从不同视角捕捉。数据集提供物体分割掩码、姿态和属性，并使用330个3D物体模型生成的合成图像进行增强。该数据集旨在解决机器人感知中的少样本物体分类问题，支持模型在元学习设置下无需重新训练即可泛化到新物体。

The FewSOL Dataset is designed for few-shot object learning in robotic environments. It consists of 336 real-world objects, each paired with 9 RGBD images captured from distinct viewpoints. The dataset provides object segmentation masks, poses and attributes, and is augmented with synthetic images generated from 330 3D object models. This dataset aims to address the few-shot object classification problem in robotic perception, supporting models to generalize to unseen objects without retraining under meta-learning settings.