mark000071/EnvShip-Bench_An_Environment-Enhanced_Benchmark_for_Short-Term_Vessel_Trajectory_Prediction

EnvShip-Bench

EnvShip-Bench is a benchmark for short-term vessel trajectory prediction built from raw AIS data released by the Danish Maritime Authority (DMA).

This Hugging Face release currently provides two source branches under a shared layout:

• DMA/
• NOAA/

The DMA branch is organized under:

• DMA/benchmark/core/
• DMA/benchmark/full/
• DMA/mini_benchmark/ship_core_lite/
• DMA/mini_benchmark/clean_ship_core_lite_v1/

The NOAA branch is organized under:

• NOAA/benchmark/core/
• NOAA/benchmark/full/
• NOAA/mini_bench/clean_ship_core_lite_v1/

The layout keeps each source in its own top-level directory so new branches can be added without changing the public repository structure.

Forecasting Protocol

All benchmark samples follow the same protocol:

• Observation horizon: 10 minutes
• Prediction horizon: 10 minutes
• Sampling interval: 20 seconds
• History length: 30 points
• Future length: 30 points

Included Data

1. DMA/benchmark/core

The main large-scale benchmark release with strict quality control for standardized vessel trajectory prediction.

2. DMA/benchmark/full

A more inclusive release that retains additional valid windows beyond the strict core subset.

3. DMA/mini_benchmark/ship_core_lite

A lightweight representative mini benchmark for quick experimentation and method prototyping.

4. DMA/mini_benchmark/clean_ship_core_lite_v1

A quality-first compact benchmark with stricter filtering and controlled motion profiles.

This subset also includes:

• environment_v1
• environment_v2
• social_env_v1

These packages support environment-aware and interaction-aware vessel forecasting on top of the same compact split.

5. DMA/data_raw/dma/incoming/2025-09/aisdk-2025-09-*.zip

The DMA branch includes the complete September 2025 daily raw files used to build the released benchmark. The original DMA CSV field note is also included in the same directory.

6. NOAA/data_raw/noaa/2025-03/ais-2025-03-*.csv.zst

The NOAA branch includes the March 2025 raw daily files used to build the released benchmark.

Quick Start

The main files are CSV-based benchmark shards:

• train/part-*.csv.gz
• val/part-*.csv.gz
• test/part-*.csv.gz

Each row is one fixed-length trajectory sample with 30 historical points and 30 future points, together with metadata and quality labels.

Reproducibility

The dataset construction code, preprocessing pipeline, visualization scripts, and paper-writing notes are maintained in the companion repository.

This dataset can be used directly for vessel trajectory forecasting, and it is also designed to support adaptation of trajectory forecasting methods originally developed for pedestrian prediction.

For model background and forecasting methodology inspiration, please refer to our paper:

P2R-Net: Prior-to-Refinement Multimodal Trajectory Forecasting for Feasible Pedestrian Motion Prediction