TL;DR本研究探讨了预训练的 ViT 和 ResNet 特征层在量化个体三维形状的二维草图视图之间的相似性方面的能力,并使用对比学习细化预训练模型,研究所选微调策略如何影响零样本形状检索准确性,提供洞察和指导采用大规模预训练模型作为感知损失的研究。
Abstract
We conduct a detailed study of the ability of pretrained on pretext tasks ViT
and ResNet feature layers to quantify the similarity between pairs of 2D sketch
views of individual 3d shapes. We assess the performance in terms of the
models' abilities to retrieve similar views and ground-
This paper explores the use of pre-trained models and synthetic renderings to generate 3D shapes from sketches without the need for paired datasets, demonstrating the effectiveness of the approach for generating multiple 3D shapes per input sketch regardless of their level of abstraction.