Our Blobby and Our Object

本研究提出了两个新的数据集：Our Blobby和Our Object，用于训练和评估光度立体技术。Our Blobby包含3000个具有平滑表面的对象，而Our Object则包含76个具有复杂几何细节的对象。这两个数据集均使用了超过1100种真实材料，以及通过随机设置参数生成的无限多种材料，以模拟非朗伯反射。数据集的创建过程利用了Blender软件的Cycles渲染引擎，确保了光照和材质的真实性。这些数据集主要应用于光度立体问题的深度学习模型训练，旨在解决复杂光照和材质条件下的3D表面重建问题。

This study proposes two novel datasets: Our Blobby and Our Object, for training and evaluating photometric stereo techniques. Our Blobby consists of 3000 objects with smooth surfaces, while Our Object contains 76 objects featuring complex geometric details. Both datasets utilize over 1100 real-world materials, as well as an infinite variety of materials generated via randomized parameter configuration, to simulate non-Lambertian reflections. The datasets are created using the Cycles rendering engine of Blender software, which ensures the authenticity of lighting and materials. These datasets are primarily applied for training deep learning models for photometric stereo problems, aiming to solve 3D surface reconstruction tasks under complex lighting and material conditions.