Closed cv-lab-x closed 3 months ago
Thank you for reporting this issue. After ensuring the camera parameters are set correctly, I have observed a noticeable improvement in the reconstruction process. Nevertheless, there are still some problem that need to be addressed. I anticipate that this endeavor will take approximately one to two weeks to complete.
xrsfm results (with given camera parameters):
The main cause of this issue is the incorrect image rotation when using cv::imread without IMREAD_IGNORE_ORIENTATION. The commit (https://github.com/openxrlab/xrsfm/commit/3439677f798e73f3ac5aeb00e730fe9aed8eb194) has already fixed this issue.
Please note that when calling the reconstruction command, the camera intrinsic parameters need to be specified.
./bin/run_reconstruction /data/person-hall/ /data/person-hall/sparse/cameras.txt /cephfs/person-hall/results/
xrsfm results:
The main cause of this issue is the incorrect image rotation when using cv::imread without IMREAD_IGNORE_ORIENTATION. The commit (3439677) has already fixed this issue.
Please note that when calling the reconstruction command, the camera intrinsic parameters need to be specified.
./bin/run_reconstruction /data/person-hall/ /data/person-hall/sparse/cameras.txt /cephfs/person-hall/results/
xrsfm results:
great, huge improvement, I'll try, thanks for your work./
thanks for your great work, xrsfm is much faster than colmap, but the results seem to be poor, i use the default params in the readme, how to improve the results? the link of the demo datasets: https://demuc.de/colmap/datasets/ Person-hall
xrsfm results:
the gt:![image](https://github.com/openxrlab/xrsfm/assets/26067407/8686dcf0-4532-442f-a0c4-fe64538bfff8)
Looking forward to your reply, thanks! @oneLOH