Accelerating the training process

[yavon818]

I wonder if it is possible to accelerate the training by introducing the cloud point as the Gaussian prior to this model.

[yavon818]

感谢您的回复，我还有一些问题想请教，就是我们试验了一些固定视角的效果，包括demo以及自己拍摄的视频，想问下哪些视频效果会好一点哪些会差一点：比方说和相机或者主体运动的方向，运动速度，背景的纹理等等

获取 Outlook for iOShttps://aka.ms/o0ukef

---

发件人: generative-dynamics @.> 发送时间: Friday, September 8, 2023 10:55:10 AM 收件人: google/dynibar @.> 抄送: yavon818 @.>; Author @.> 主题: Re: [google/dynibar] Accelerating the training process (Issue #36)

I think this is a very interesting research direction!

― Reply to this email directly, view it on GitHubhttps://github.com/google/dynibar/issues/36#issuecomment-1711005988, or unsubscribehttps://github.com/notifications/unsubscribe-auth/BB45IU2BIUVROA3UKD56LUTXZKCI5ANCNFSM6AAAAAA4KIO3DI. You are receiving this because you authored the thread.Message ID: @.***>

[yavon818]

我还注意到一个固定视角的视频（就是绿衣女子在树下向前奔跑），最开始几帧胳膊很模糊，缺了部分，在我们自己拍的人向前走，相机向后移动也出现了这个情况，想问下这个是如何导致的，可以优化吗

获取 Outlook for iOShttps://aka.ms/o0ukef

---

发件人: ZHANG Yawen @.> 发送时间: Friday, September 8, 2023 4:47:51 PM 收件人: google/dynibar @.>; google/dynibar @.> 抄送: Author @.> 主题: Re: [google/dynibar] Accelerating the training process (Issue #36)

感谢您的回复，我还有一些问题想请教，就是我们试验了一些固定视角的效果，包括demo以及自己拍摄的视频，想问下哪些视频效果会好一点哪些会差一点：比方说和相机或者主体运动的方向，运动速度，背景的纹理等等

获取 Outlook for iOShttps://aka.ms/o0ukef

---

发件人: generative-dynamics @.> 发送时间: Friday, September 8, 2023 10:55:10 AM 收件人: google/dynibar @.> 抄送: yavon818 @.>; Author @.> 主题: Re: [google/dynibar] Accelerating the training process (Issue #36)

I think this is a very interesting research direction!

― Reply to this email directly, view it on GitHubhttps://github.com/google/dynibar/issues/36#issuecomment-1711005988, or unsubscribehttps://github.com/notifications/unsubscribe-auth/BB45IU2BIUVROA3UKD56LUTXZKCI5ANCNFSM6AAAAAA4KIO3DI. You are receiving this because you authored the thread.Message ID: @.***>

[yavon818]

看到了补充材料中的“the performance of our method degrades if the target viewpoint is from an oblique angle with respect to the input views”. 不是很明白这里的oblique是指什么，是视角的translation还是rotation

获取 Outlook for iOShttps://aka.ms/o0ukef

---

发件人: ZHANG Yawen @.> 发送时间: Friday, September 8, 2023 4:50:32 PM 收件人: google/dynibar @.>; google/dynibar @.> 抄送: Author @.> 主题: Re: [google/dynibar] Accelerating the training process (Issue #36)

我还注意到一个固定视角的视频（就是绿衣女子在树下向前奔跑），最开始几帧胳膊很模糊，缺了部分，在我们自己拍的人向前走，相机向后移动也出现了这个情况，想问下这个是如何导致的，可以优化吗

获取 Outlook for iOShttps://aka.ms/o0ukef

---

发件人: ZHANG Yawen @.> 发送时间: Friday, September 8, 2023 4:47:51 PM 收件人: google/dynibar @.>; google/dynibar @.> 抄送: Author @.> 主题: Re: [google/dynibar] Accelerating the training process (Issue #36)

感谢您的回复，我还有一些问题想请教，就是我们试验了一些固定视角的效果，包括demo以及自己拍摄的视频，想问下哪些视频效果会好一点哪些会差一点：比方说和相机或者主体运动的方向，运动速度，背景的纹理等等

获取 Outlook for iOShttps://aka.ms/o0ukef

---

发件人: generative-dynamics @.> 发送时间: Friday, September 8, 2023 10:55:10 AM 收件人: google/dynibar @.> 抄送: yavon818 @.>; Author @.> 主题: Re: [google/dynibar] Accelerating the training process (Issue #36)

I think this is a very interesting research direction!

― Reply to this email directly, view it on GitHubhttps://github.com/google/dynibar/issues/36#issuecomment-1711005988, or unsubscribehttps://github.com/notifications/unsubscribe-auth/BB45IU2BIUVROA3UKD56LUTXZKCI5ANCNFSM6AAAAAA4KIO3DI. You are receiving this because you authored the thread.Message ID: @.***>