search

JonathonLuiten / Dynamic3DGaussians

Other
1.82k stars 107 forks source link

point cloud and image size #24

Open luoxue-star opened 6 months ago

luoxue-star commented 6 months ago

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 commented 2 months ago

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 commented 2 months ago

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

JHXion9 commented 2 months ago

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! face

luoxue-star commented 2 months ago

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