Training Sequences
Test Sequences

LaMAria is a city-scale egocentric SLAM dataset that provides sequences captured in various challenging environments. We offer 23 training and 63 test sequences. The training and test sequences include sequence and calibration data in the formats discussed in the documentation. The sparse and dense pseudo ground truth files are provided for the training sequences and kept closed for the test sequences. You can submit your results for both training and test sequences for evaluation via this website.

Download Format

We provide a Python script download_lamaria.py for convenience in data download. The script downloads and organizes the data in the following directory structure:

    output_dir/
    ├── training/
    │   ├── <sequence_name>/
    │   │   ├── aria_calibrations/
    │   │   │   └── <sequence_name>.json
    │   │   ├── asl_folder/
    │   │   │   ├── <sequence_name>.zip
    │   │   ├── ground_truth/
    │   │   │   ├── pGT/
    │   │   │   │   └── <sequence_name>.txt
    │   │   │   └── control_points/
    │   │   │       └── <sequence_name>.json
    │   │   ├── pinhole_calibrations/
    │   │   │   └── <sequence_name>.json
    │   │   ├── raw_data/
    │   │   │   └── <sequence_name>.vrs
    │   │   └── rosbag/
    │   │       └── <sequence_name>.bag
    │   └── ....
    └── test/
        ├── <sequence_name>/
        │   └── # no ground truth available
        └── ...

The tables below summarize the dataset contents. To watch a sped-up, downscaled preview of each sequence, click on the sequence name. Individual sequence data can be downloaded from the links given below.

Training Sequences

Test Sequences