在深度学习中准备训练数据这是一种什么技术？没见过 求解
论文：GeoDesc: Learning Local Descriptors by Integrating Geometry Constraints
”we rely on successful 3D reconstructions to generate ground truth 2D correspondences in an auto-matic manner. First, sparse reconstructions are obtained from standard SfMpipeline [32]. Then, 2D correspondences are generated by projecting 3D point clouds.”
https://www.zhihu.com/question/29885222?sort=created
先通过几幅得到的图像，进行稀疏三维重建获得三维点云模型，再对三维点云进行任意角度二维映射吧，这样可以得到任意角度下的任意多幅二维图像，从而生成充分的训练数据，应该是这个意思吧（如果楼主满意，请打个赏，1个也行，我要下个东西，谢谢）