Dora-bench
收藏资源简介:
Dora-bench是一个用于评估3D变分自编码器(VAE)重建质量的基准数据集。该数据集通过几何复杂度对测试形状进行分类,分为四个级别:较少细节、中等细节、丰富细节和非常丰富细节。数据集整合了来自ABO、GSO、Meta和Objaverse等多个公共数据集的3D形状,旨在通过新的Sharp Normal Error(SNE)指标,更严格地评估3D VAE在重建精细几何特征方面的表现。该数据集的应用领域主要集中在3D内容生成和建模,旨在解决现有3D VAE在几何细节捕捉和重建精度上的不足。
Dora-bench is a benchmark dataset for evaluating the reconstruction quality of 3D variational autoencoders (VAEs). This dataset categorizes test shapes by geometric complexity into four levels: Low Detail, Medium Detail, High Detail, and Ultra High Detail. It incorporates 3D shapes from multiple public datasets including ABO, GSO, Meta, and Objaverse, aiming to conduct more rigorous evaluation of 3D VAEs' performance in reconstructing fine-grained geometric features via the novel Sharp Normal Error (SNE) metric. This dataset is primarily applied in 3D content generation and modeling, with the goal of addressing the shortcomings of existing 3D VAEs in geometric detail capture and reconstruction accuracy.




