DVS-PedX: Dynamic Vision Sensor Pedestrian eXploration Dataset

DVS-PedX (Dynamic Vision Sensor Pedestrian eXploration Dataset)

DVS-PedX is a neuromorphic dataset designed for pedestrian detection and crossing-intention analysis in normal and adverse weather conditions across two complementary sources: (1) synthetic event streams generated in the CARLA simulator for controlled “approach–cross” scenes under varied weather and lighting; and (2)real-world JAAD dash-cam 346 videos converted to event streams using the v2e tool, preserving natural behaviors and backgrounds. Event-based Pedestrian Interaction Clips (JPG/PNG + AEDAT/AEDAT4 + AVI previews).

Data Structure

DVS_PedX/├── carla_simulator/│   ├── frames_dataset_rgb/│   │   └── <sample_id>/                 # UUID-like folder│   │       └── <frame_index>-<label>.jpg   # label: 0=no crossing, 1=crossing│   ├── frames_dataset_dvs/│   │   └── <sample_id>/│   │       └── <frame_index>-<label>.png│   ├── frames_dataset_weather_rgb/│   │   └── <sample_id>/  → <frame_index>-<label>.jpg│   └── frames_dataset_weather_dvs/│       └── <sample_id>/  → <frame_index>-<label>.png│└── JAAD_DVS/    ├── video_0001/    │   ├── video_0001_v2.aedat            # AEDAT v2 event stream    │   ├── video_0001_v4.aedat4           # AEDAT v4 event stream    │   ├── dvs-video.avi                  # preview video    │   ├── dvs-video-frame_times.txt      # preview frame timestamps    │   └── v2e-args.txt                   # v2e conversion parameters    ├── video_0002/    └── ...

DVS_PedX Sample Dataset provides a lightweight, representative subset of the full DVS_PedX benchmark, designed for demonstration purposes.

The sample includes some sequences from both CARLA and JAAD event-based pedestrian datasets, covering Dynamic Vision Sensor (DVS) data and corresponding modalities where available (e.g., RGB and adverse-weather variants in CARLA). All files preserve the original directory structure and file formats of the full datasets.

Overview

Starting from v2, the DVS-PedX dataset is distributed as single compressed archives (.tar.xz) instead of multi-part files and contains the Samples dataset from Carla and JAAD.

This simplifies downloading, verification, and extraction, and reduces the risk of file corruption. Each archive is accompanied by a SHA-256 checksum file for integrity verification.

Dataset Download and Extraction Instructions (v2)

Step 1 — Download

Download the desired archive(s) and the corresponding .sha256 file from Zenodo, for example:

DVS_PedX_CARLA_FULL.tar.xzDVS_PedX_CARLA_FULL.tar.xz.sha256

Step 2 — Verify integrity (optional, but recommended)

macOS / Linux

shasum -a 256 -c DVS_PedX_*.tar.xz.sha256

Windows (PowerShell)

Get-FileHash DVS_PedX_XXXX.tar.xz -Algorithm SHA256

*Compare the output with the value in the .sha256 file.

Step 3 — Extract

macOS / Linux

tar -xJvf DVS_PedX_XXXX.tar.xz

Windows

Extract using 7-Zip (right-click → Extract Here).

Step 4 — Use

Use the dataset directly from the extracted directory (e.g., DVS_PedX_CARLA_FULL/).

We don't recommend modifying the internal directory structure.

Dataset preparation and sample splitting scripts are publicly available at https://github.com/AVHeros/DVS-PedX-DataSplitter.