Loping151 / ForPlane

Neural LerPlane Representations for Fast 4D Reconstruction of Deformable Tissues
Other
48 stars 2 forks source link

About average_poses function? #9

Closed xuyaojian123 closed 4 months ago

xuyaojian123 commented 11 months ago

I've always had a doubt. Could you explain why normalize(np.cross(z, y_)) is not -x?The coordinate system of poses is "right up back". Looking forward your reply. image

Loping151 commented 11 months ago

(首先,这个函数并没有被我们用到。。不过你似乎在kplane也问了。。 (抱歉晚回复,前几天在考试。。 看的我一开始也很疑惑,但是 image 如果你认为他的字母都是标在正方向上的,那似乎的确是左手系。

xuyaojian123 commented 11 months ago

haha,谢谢你哈!

xuyaojian123 commented 11 months ago

想请问下你的采样方式是用K-planes的proposal sample方法,还是NerfAcc工具包提供的采样呢?

Loping151 commented 11 months ago

我写了个参数控制采样方法,不过不太直观(my bad config中调整参数occ_grid_resoself.use_occ_grid = self.occ_grid_reso > 0)控制使用proposal plane来sample还是nerfacc的occupancy grid

xuyaojian123 commented 11 months ago

好的,谢谢。一般来说你是觉得nerfacc的occupancy grid的sample方法效果好点,还是proposal plane来sample的方法效果好点?

Loping151 commented 11 months ago

我认为,就效果而言对于我们的任务差别并不显著,但是grid的方法的效果在于离散化了采样步骤,inference(即渲染速度)够快。对于一些更大的场景,比如尝试结合像zipnerf那样的工作,具体效果有待研究

xuyaojian123 commented 11 months ago

好的,谢谢。能不能加个wx([REDACTED]),我有几个mip-nerf论文的问题想请教你一下,周围没人可以请教的哈哈哈!!!