Tools

• summary table and plots: python -m trajnetplusplustools.summarize <dataset_files>
• plot sample trajectories: python -m trajnetplusplustools.trajectories <dataset_file>
• visualize interactions: python -m trajnetplusplustools.visualize_type <dataset_file>
• obtain distribution of trajectory types: python -m trajnetplusplustools.dataset_stats <dataset_file>

APIs

• trajnetplusplustools.Reader: class to read the dataset_file
• trajnetplusplustools.show: module containing contexts for visualizing rows and paths
• trajnetplusplustools.writers: write a trajnet dataset file
• trajnetplusplustools.metrics: contains unimodal metrics: average_l2(), final_l2() and collision() and multimodal metrics: topk() and nll() implementations

Dataset

Datasets are split into train, val and test set. Every line is a self contained JSON string (ndJSON).

Scene:

{"scene": {"id": 266, "p": 254, "s": 10238, "e": 10358, "fps": 2.5, "tag": 2}}

Track:

{"track": {"f": 10238, "p": 248, "x": 13.2, "y": 5.85}}

with:

• id: scene id
• p: pedestrian id
• s, e: start and end frame id
• fps: frame rate
• tag: trajectory type
• f: frame id
• x, y: x- and y-coordinate in meters
• pred_number: (optional) prediction number for multiple output predictions
• scene_id: (optional) corresponding scene_id for multiple output predictions

Frame numbers are not recomputed. Rows are resampled to about 2.5 rows per second.

Dev

pylint trajnetplusplustools
python -m pytest
# optional: mypy trajnetplusplustools --disallow-untyped-defs

Dataset Summaries

biwi_hotel:

crowds_students001:

crowds_students003:

crowds_zara02:

crowds_zara03:

dukemtmc:

syi:

wildtrack:

Interactions

leader_follower:

collision_avoidance:

group:

others:

Citation

If you find this code useful in your research then please cite

@inproceedings{Kothari2020HumanTF,
  title={Human Trajectory Forecasting in Crowds: A Deep Learning Perspective},
  author={Parth Kothari and Sven Kreiss and Alexandre Alahi},
  year={2020}
}