Helpers for pulling down manageable slices of the Argoverse 2 LiDAR dataset and poking at the sweeps/poses without touching the rest of the repo.

• download_sample_logs.py – Enumerates S3 log prefixes and syncs a limited count per split into ../argverse_data_preview, forwarding concurrency/part-size tweaks to s5cmd.
• preview_lidar_gui.py – Tkinter picker plus Open3D viewer for LiDAR sweep .feather files, with optional root/stride overrides and a pose picker to render sweeps in city coordinates.
• preview_city_pose_gui.py – Table view of city_SE3_egovehicle.feather data with summary stats (duration, cadence, translation bounds).
• plot_pose_on_map.py – Converts pose translations from city ENU to WGS84 via the published anchors and renders the trajectory on an OpenStreetMap basemap.
• export_pose_coordinates.py – Streams every pose file from the Argoverse S3 bucket and appends the results to av2_coor.feather.
• aggregate_heatmap_cells.py – Bins poses into 500m grid cells with counts and per-cell opacity suggestions, writing city_heatmap_cells.json.
• compute_city_bounds.py, render_city_bounds_map.py, city_bounds.json, city_bounds_map.html – Extract city extents and visualise their rectangles.
• render_city_heatmap_map.py, city_heatmap_map.html – Overlay the city bounds and heatmap cells with transparency scaling to highlight density.
• av2_coor.feather – Materialised pose table (~0.9 GB) with quaternion, translation, UTM, and WGS84 coordinates for every log/timestamp in the LiDAR splits.
• environment.yml, requirements.txt – Dependency manifests (numpy, pandas, pyarrow, open3d, pyproj, folium, etc.).
• s5cmd/ – Bundled Windows binary (s5cmd.exe) plus upstream CHANGELOG, LICENSE, and README for high-throughput S3 transfers.

Create an environment inside this folder with either conda or pip:

conda env create -f environment.yml
conda activate argoverse-lidar

or

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

download_sample_logs.py lists the S3 prefixes under s3://argoverse/datasets/av2/lidar/ and syncs a limited number of logs per split via the bundled s5cmd binary. By default it fetches 100 validation logs into ..\argverse_data_preview\val, auto-creating the destination if needed.

Key flags:

• --split repeatable; defaults to just val.
• --count number of logs per split to mirror (default 100).
• --dest target directory (default ..\argverse_data_preview).
• --skip-existing leaves already-downloaded logs untouched.
• --concurrency, --part-size, and --extra-arg are forwarded to s5cmd for tuning throughput.
• --s5cmd overrides the path to the executable if you have your own build.

Example command:

python download_sample_logs.py --split train --split val --count 25 --skip-existing

The script streams the s5cmd output so you can keep an eye on progress and abort safely with Ctrl+C.

• preview_lidar_gui.py opens a Tkinter list of .feather sweeps and renders the selected cloud in Open3D. Use --root to point at a custom folder and --stride to subsample points, or supply a log's city_SE3_egovehicle.feather via the pose picker to stack sweeps in city coordinates.
• preview_city_pose_gui.py loads a city_SE3_egovehicle.feather file, summarises timestamps/translations, and shows up to 500 rows in a table. Launch it with python preview_city_pose_gui.py --file <path> or use the built-in file picker.
• plot_pose_on_map.py converts pose translations to WGS84 using the published anchors and writes an interactive HTML map. Pass --pose <path> plus optional --city, --output, and --no-open flags.

• export_pose_coordinates.py streams every city_SE3_egovehicle.feather from the public Argoverse bucket and writes a consolidated table to av2_coor.feather.
• The exported dataset now resides at ArgoverseLidar/av2_coor.feather (~0.9 GB, ~6 M rows) and is ready for downstream analytics.
• Each row represents a timestamped ego pose with both UTM and WGS84 projections derived via the city anchors.
• Per-log pose files (city_SE3_egovehicle.feather) keep timestamp_ns aligned with a quaternion (qw, qx, qy, qz) and translation in metres (tx_m, ty_m, tz_m) for each sweep.

Pose table columns:

• city, split, log_id – Provenance of each measurement.
• timestamp_ns – Capture time in nanoseconds.
• qw, qx, qy, qz – Ego pose quaternion.
• tx, ty, tz – Ego translation in the city frame (metres).
• zone_num_hemi – UTM zone label (all northern hemisphere).
• easting, northing – Projected UTM coordinates in metres.
• latitude, longitude – WGS84 coordinates computed via the published city anchors.

Sample rows (first 5):

• aggregate_heatmap_cells.py buckets the pose table into 500 m x 500 m UTM cells per city and records the point density (city_heatmap_cells.json).
• compute_city_bounds.py and render_city_bounds_map.py create a quick bounds overlay (city_bounds.json → city_bounds_map.html).
• render_city_heatmap_map.py layers the per-city rectangles and the heatmap cells onto OpenStreetMap (city_heatmap_map.html), modulating opacity between 0.1–0.8 based on counts.

The s5cmd/ folder contains s5cmd.exe (v2.3.0) and upstream documentation so Windows users do not need to install it separately. The downloader script points to this binary by default but accepts a custom path when required.