CookMaker

对于问题1，我已经大概认定这是位姿，但是对于原视频一眨眼，推理的人脸就忽大忽小这个问题我还是比较困惑

> > 你可以增加landmark的平滑窗口，一定程度上可以改善忽大忽小的问题 > > 你说的是，在预估的时候infer_smooth_camera_path_kernel_size么？ 我的做法是增加训练时的哪个smo_win_size，那个是用来平滑landmark的

Same problem,the hair cannot be represented well.

Hello, I met the same problem this morning. You can fine the answer in #24 .Hope it could solve your problem.

By the way,the nan loss leads to a totally wrong result,you can check your tensorboard to find a checkpoint that can be use.

just modified the file "--config=egs/datasets/May/lm3d_radnerf_sr.yaml",find the ”ambient_loss_target“ and increase the value

Same question

> Hi [@Isabella98Liu](https://github.com/Isabella98Liu) , thanks for your great work. > > I wonder how you extract the mesh from those NeRF-based methods? Just like possion reconstuction from depth map in...

instant-ngp in my work is just a accelerate tool and i did not break the framework,all things works fine(in my opinion...)

oh! I realized that the tensorboard result is rendered by only a few input images. I got a better result after 4500 iters:  And i will try to render...