VISION-SJTU / Lightning-NeRF

[ICRA 2024] Lightning NeRF: Efficient Hybrid Scene Representation for Autonomous Driving
MIT License
105 stars 6 forks source link

refine depth #7

Open liucsg opened 4 months ago

liucsg commented 4 months ago

lightning nerf 优化得到的depth,进行几何重建,效果会比初始的Point cloud 更优吗?

Lzc6996 commented 4 months ago

@liucsg 并不会,nerf的重点是稠密重建和多角度渲染。

liucsg commented 1 month ago

请教一个问题,如果用带lidar的点云通过pose,累积输出场景几何,可能有小部分区域有缺失,然后渲染各个视角下的图片,那么也就能拿到渲染图片下的depth,去做mvs训练会有收益吗?另外,请教一下,lightning-nerf这一方向,可以缓解不同车不同安装位置相机带来的ground-truth的难题吗?期待您的回复,感谢!

---- 回复的原邮件 ---- | 发件人 | Zhichao @.> | | 日期 | 2024年05月09日 21:18 | | 收件人 | @.> | | 抄送至 | @.>@.> | | 主题 | Re: [VISION-SJTU/Lightning-NeRF] refine depth (Issue #7) |

@liucsg 并不会,nerf的重点是稠密重建和多角度渲染。

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

Lzc6996 commented 1 month ago

@liucsg mvs通常都会有收益。lightning nerf具备一定的外插能力,能一定程度的缓解硬件位置区别带来的影响。