Miniecho/EVMars-Anomaly-v2

---
language:
- en
pretty_name: Anomaly Detection
size_categories:
- 100M<n<1B
---

# EVMars-Anomaly-v2: A Multimodal Color Event Dataset for Martian Anomaly Detection and Dust Storm Estimation

**EVMars-Anomaly-v2**, an updated and expanded version of the original EVMars-Anomaly dataset (a specialized subset of EVMars), remains a multimodal dataset dedicated to anomaly detection in simulated Martian dust storm scenarios using a color event camera. This version introduces significant enhancements to address sensing challenges in extremely low-visibility environments, thereby supporting more reliable particulate matter estimation and anomaly detection under severe dust storm conditions.

The dataset consists of 68 sequences captured with a DAVIS346 color event camera(346×260 resolution, 1 µs temporal precision) in a outdoor data collection test field. High-power industrial fans simulate three canonical dust storm intensities:

- **Local Dust Storm**: Mild particle suspension
- **Regional Dust Storm**: Moderate occlusion
- **Global Dust Storm**: Extreme low-visibility

The data format is as follows:

Event Streams: Stored as CSV files with the following columns:
timestamp (microseconds), x (pixel column, 0–345), y (pixel row, 0–259), polarity (±1), color (RED/GREEN/BLUE).
Each file corresponds to one continuous sequence.

Ground Truth: Provided in a separate synchronized CSV per sequence:
timestamp (ms), TSP (µg/m³), PM2.5 (µg/m³), PM10 (µg/m³)

## Key Updates in v2 (Released January 2026)

- Added numerous new sequences with substantially higher dust concentrations, enabling better modeling and evaluation of performance in extreme dust storm environments.
- Expanded the total number of event data points to approximately 680 million, providing richer temporal and spatial event distributions.
- Introduced standardized dataset splits divided into three dust storm intensity intervals: **global** (planet-encircling storms with highest dust loading), **regional** (large-scale but localized storms), and **local** (smaller, confined dust events). This partitioning facilitates consistent benchmarking and analysis across varying severity levels.

## Historical Version

**EVMars-Anomaly-v1**: The original release containing sequences covering a baseline range of dust conditions and approximately 150 million event data points.Available at: [Hugging Face Dataset v1](https://huggingface.co/datasets/Miniecho/EVMars-Anomaly)

## License
This repository is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).