multipleviewprogress.sh failure to

[liuhaixu2021]

I attempted to adapt CMU's Panoptic dataset to a multi-view dataset. However, when running multipleviewprogress.sh to generate point clouds via colmap using 10 viewpoints (24 frames per viewpoint), I encountered errors in all five scenarios with the message “No good initial image pair found.” I'd like to ask what might be causing this issue. I used the default parameters in the sh file—could the default settings be too strict? Or is it possible that even when “No good initial image pair found” occurs, the best parameter combination can still be found in the db file to continue executing the code?

[CarlDegio]

I encountered the same problem, which I believe is caused by colmap's inability to utilize sparse viewpoints. This error usually occurs when colmap reconstruction doesn't effectively utilize all camera images. You can try reconstructing using the colmap GUI and find that only a small number of cameras are recognized.

Sparse reconstruction methods might be better. I've currently tested VGGT (which has a conversion to Colmap format), but it's not particularly good either.

[Harryhaozhu]

我也遇到了同样的问题，我认为这是由于 colmap 无法有效利用稀疏视点造成的。当 colmap 重建未能有效利用所有相机图像时，通常会发生此错误。您可以尝试使用 colmap 图形用户界面进行重建，并发现只有少数相机被识别。

稀疏重建方法或许更好。我目前测试了 VGGT（它可以转换为 Colmap 格式），但效果也并非特别理想。

您好，我使用cut_roasted_beef数据集测试多视角训练，multipleviewprogress.sh倒是成功了，但是训练的结果特别差，非常的混乱，您知道这是哪里的问题吗，与一开始的输入数据有关吗