1,000 object classes containing 1.3 million event-based recordings derived from the original ImageNet dataset. The data consists of asynchronous event streams captured by moving a 640x480 resolution event camera in front of a monitor displaying static images to simulate real-world motion.
Use Cases
- Train spiking neural networks (SNNs) for high-speed object recognition using the polarity and timestamp features
- Develop event-to-frame reconstruction algorithms using the asynchronous event streams and class labels
- Benchmark the noise-tolerance of event-based vision algorithms using the raw (x, y, t, p) data
Strengths
- 1,000 object categories following the ILSVRC-2012 hierarchy
- 1.3 million samples recorded as (x, y, timestamp, polarity) event streams
- Captured using a 640x480 resolution event-based sensor