CVPR 2019 Self-Supervised Learning of 3D Human Pose using Multi-view Geometry 算法研究
step1inputnnn个相机(n≥2n≥2n≥2),同时对场景中的人拍照,连续相机的baseline较小https://blog.csdn.net/zhang11wu4/article/details/41827073产生I1,I2,...InI_1,I_2,...I_nI1,I2,...In张图片,相邻的图像两两产生图像对,就是训练集training_data:{(Ii,Ii+1)∣i=1,