search

cvg / pixloc

Back to the Feature: Learning Robust Camera Localization from Pixels to Pose (CVPR 2021)
Apache License 2.0
735 stars 92 forks source link

How the intrinsic values are calculated for Cambridge and 7 Scenes dataset? #19

Closed patelajaychh closed 2 years ago

patelajaychh commented 2 years ago

Intrinsic parameters for query images in Cambridge and 7Scenes dataset are not available. But PixLoc requires calibrated query images so how these parameters are calculated to evaluating on these datasets?

Thanks!

sarlinpe commented 2 years ago

The intrinsics are actually available. They are estimated from SfM as part of the ground truthing process. Here we simply extract them from the GT SfM models using hloc: https://github.com/cvg/Hierarchical-Localization/blob/master/hloc/pipelines/Cambridge/utils.py#L42 The GT SfM models were kindly provided by the authors of the recent ICCV 2021 paper On the Limits of Pseudo Ground Truth in Visual Camera Re-localisation. Checkout the hloc pipelines to see how we processed the raw data: https://github.com/cvg/Hierarchical-Localization/tree/master/hloc/pipelines/Cambridge

patelajaychh commented 2 years ago

Thanks! Can you please also provide GT poses for Cambridge and 7Scenes used for evaluating PixLoc paper.

sarlinpe commented 2 years ago

The COLMAP models containing the GT query poses can be downloaded with our download.py script or with the commands linked above:

Cambridge: https://drive.google.com/file/d/1esqzZ1zEQlzZVic-H32V6kkZvc4NeS15/view 7Scenes: https://drive.google.com/file/d/1cu6KUR7WHO7G4EO49Qi3HEKU6n_yYDjb/view

patelajaychh commented 2 years ago

I'm running SFM reconstruction of Cambridge KingsCollege scene using command- colmap mapper --database_path /data/hloc/outputs_KingsCollege/sfm_superpoint+superglue/database.db --image_path /data/datasets/Cambridge/KingsCollege/images_all --output_path /data/hloc/outputs_KingsCollege/sfm_superpoint+superglue/models --Mapper.num_threads 16

There are total 1565 images. Its already 8hr and its still running. Is this usual? Also process is not using GPU. Does mapper function in COLMAP not support GPU?

I'm wondering how much time did it take to create above shared ground truths? Is it possible to share time profile of each scene?