search

TQTQliu / MVSGaussian

[ECCV 2024] MVSGaussian: Fast Generalizable Gaussian Splatting Reconstruction from Multi-View Stereo
https://mvsgaussian.github.io/
MIT License
416 stars 21 forks source link

Build self dataset problem when colmap generate pose #64

Open 5p6 opened 1 month ago

5p6 commented 1 month ago

When I tried to generate the MVSGaussian dataset using the video frame data I collected myself, I encountered some problems. When I run the command

python lib/colmap/imgs2poses.py -s examples/myscene

It always reports information

Need to run COLMAP
Features extracted
Features matched
Sparse map created
Finished running COLMAP, see .\dataset\colmap_output.txt for logs
Post-colmap
Cameras 5
Images # 21
ERROR: the correct camera poses for current points cannot be accessed
Done with imgs2poses

It seems that there is a problem with the camera pose and corresponding point cloud,But the most crucial thing is that it cannot generate the data poses_bounds.npy to be used in the next stage. Can you provide some suggestions for me to collect datasets that are compatible with MVSGaussian,for example, the parameters captured by the camera, as well as the characteristics of the images used for processing, such as image size, image quality, and image clarity.

TQTQliu commented 1 month ago

Hi, you can try runing colmap sparse reconstruction https://colmap.github.io/cli.html# , and then run the script lib/colmap/imgs2poses.py to get poses_bounds.npy.

datouzhan commented 1 week ago

This is because there are images in your image that cannot estimate the camera pose. It is recommended to manually filter the following images. it works