Image Matching: Local Features & Beyond 2020

This page holds links to the Image Matching Challenge PhotoTourism (IMC-PT) 2020 dataset.

For more details see https://image-matching-challenge.github.io/

Training Data

All files are compressed as tar.gz files

• brandenburg_gate (4.0G)
• buckingham_palace (5.2G)
• colosseum_exterior (6.4G)
• grand_place_brussels (3.6G)
• notre_dame_front_facade (13G)
• palace_of_westminster (3.2G)
• pantheon_exterior (4.6G)
• taj_mahal (4.2G)
• temple_nara_japan (3.0G)
• trevi_fountain (11G)

*Note:

• We further removed the hagia_sophia_interior, westminster_abbey in 2022 due to data inconsistencies.
• We removed the prague_old_town in 2021 due to data inconsistencies.

Validation Data

The validation data is composed of the following sequences: sacre_coeur, st_peters_square, reichstag.

• all sequences

sacre_coeur, st_peters_square and reichstag are also provided in the training data format separately so that they could be used as part of training if one wishes to do so.

• sacre_coeur (3.9G)
• st_peters_square (8.1G)
• reichstag (0.2G)

Test Data

• all sequences (images only, used for 2019, 2020, and 2021 IMC challenge)

We released full Phototourism test set with ground truth in 2022. The test set for 2019, 2020, and 2021 IMC challenge is sampled from full test set to have 100 images per scene.

• british_museum (2.0G)
• florence_cathedral_side (0.4G)
• lincoln_memorial_statue (2.5G)
• milan_cathedral (0.4G)
• mount_rushmore (0.4G)
• piazza_san_marco (0.8G)
• sagrada_familia (1.4G)
• st_pauls_cathedral (2.0G)