point cloud and image size

[luoxue-star]

Does the input image have to be sized at (640, 320)? I have also utilized Colmap to reconstruct the point cloud, and the data has been processed into the same format as the provided dataset. I use 13 cameras for training and the PSNR is approximately 25 at each step of the training process. However, the size of the saved params.npz file is only a few KB. What might be causing this issue? I hope to hear from you soon. Thanks!

[JHXion9]

I meet the same problems, but I solved it by using the w2c which are provided by colmap. Although I can achieve the 3D tracking with 30-psnr, the result of gaussian is so bad.

[luoxue-star]

I solved this problem by creating a point cloud of the initial frame through 3D Gaussian Splatting. @JHXion9

[JHXion9]

Can you use your own data set to get a good tracking effect, as demonstrated in this paper? I only used the loss of im, rigid, rot and iso. Although I could achieve PSNR of 30+, the tracking effect was not very good as shown in the figure. Do you have a good solution?

best!

[luoxue-star]

Sorry, I don't track well using my own dataset. This may require further strengthening of the constraint on the Gaussian between frames.

[MikeAiJF]

Hello, do you need to change the train_meta.json file to train your own data? Do you need to change the init_pt_cld.npz (specifically how to change it)? You can still use the author's.

[MikeAiJF]

我也遇到了同样的问题，不过用 colmap 提供的 w2c 解决了，虽然能实现 30-psnr 的 3D 跟踪，但是高斯的效果太差了。

我也遇到了同样的问题，不过用 colmap 提供的 w2c 解决了，虽然能实现 30-psnr 的 3D 跟踪，但是高斯的效果太差了。

为什么我训练自己的数据却显示的是作者的可视化界面，请问是什么问题

[MikeAiJF]

是需要修改train_meta.json和init_pt_cld.npz文件吗

[tianty]

@luoxue-star @JHXion9
@JHXion9 @luoxue-star Hi，I am trying to build my own dataset for training, especially for face reconstruction. My dataset has achieved good results on the vanilla Gaussian splatting, but I still can't train the initial frame on the Dynamic3DGaussians. Regarding the construction of the dataset, can you share the code you used? It would be best if there is sample data as a reference. Thank you very much!

I also tried the code in this link(#18), but still couldn't successfully build a dataset that can be used with dynamic Gaussian. When training to a certain stage, all Gaussians will be removed and only empty files will be output.