SAED: Structure-Aware Event Dataset for Neuromorphic Vision Reliability

SAED (Structure-Aware Event Dataset) is a dataset of approximately 27,000 event clips captured with a DVXplorer event camera under controlled indoor conditions. Created by Haiyu Li and accepted for publication at IJCNN 2026, it focuses on structural variation to analyze reliability in neuromorphic vision systems. The dataset includes standardized degradation protocols like noise injection and structural removal.

Use Cases

• Reliability analysis of neural models based on controlled structural degradation protocols.
• Robustness evaluation for event-based vision systems based on varied spatial scales and motion conditions.
• Structure-aware learning for neuromorphic hardware based on geometry-driven objects with clear edges.
• Development of diagnostic metrics for silent failures based on decoupled structural corruption.

Strengths

• Approximately 27,000 event clips provide a substantial basis for systematic study.
• 78.6 GB size indicates a large-scale, high-resolution dataset.
• Includes standardized structural degradation protocols (noise injection, over-denoising, localized removal) for controlled experiments.
• Organized across object geometry, spatial scale (0.5×, 1.0×, 1.5×), and motion conditions for comparability.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data is captured under controlled indoor conditions, which may limit generalizability to real-world, uncontrolled environments.

Provenance

Source

Haiyu Li

Collection Method

Captured using a DVXplorer event camera (640×480 resolution) under controlled indoor conditions.

Time Range

null

Freshness

Last updated 2026-04-13 16:57:16; freshness should be verified.

Geography

null