search

QVPR / Patch-NetVLAD

Code for the CVPR2021 paper "Patch-NetVLAD: Multi-Scale Fusion of Locally-Global Descriptors for Place Recognition"
MIT License
510 stars 72 forks source link

Groundtruths and details of the datasets used #68

Closed Anuradha-Uggi closed 1 year ago

Anuradha-Uggi commented 1 year ago

Dear Sir, You have done a great job. The current Github includes ground truths for Nordland, Pittsburgh 30k, and Tokyo 24/7. It would be helpful for us if you could provide the datasets and the ground truth npz files/links for RobotCar Seasons v2, Extended CMU Seasons, and Mapillary Street Level Sequences (MSLS). Thank you very much.

StephenHausler commented 1 year ago

Hi @Anuradha-Uggi, Unfortunately there are no publicly accessible GT files for the test sets of RobotCar Seasons v2 and Extended CMU Seasons, as these are challenge sets and require submission to the server on VisualLocalization.net, in order to generate the results. For MSLS, please follow these instructions to generate the ground truth: https://github.com/QVPR/Patch-NetVLAD/blob/main/patchnetvlad/dataset_gt_files/README_how_to_calc_mapillary_gt.txt

Closing for now, but feel free to re-open if you have any more questions.