Differentiable Volumetric Rendering: Learning Implicit 3D Representations without 3D Supervision
<DVR>
Differentiable volumetric rendering: Learning implicit 3d representations without 3d supervision目录
编者按
- 思路：首先手动推导出每个camera ray和隐表面交点的点坐标对网络参数的梯度，在实际计算时，就可以先在camera ray上采样得出交点坐标(类似二分法)，然后代入所手动推导出的式子构成完整的反向传播链路