Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -11,6 +11,37 @@ tags:
|
|
| 11 |
|
| 12 |
# EventRain-27K
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
- **Synthetic datasets (KITTI & SPAC).** Event streams are partitioned into contiguous windows of Δt = 0.1 s (10^8 ns per segment) and stored in `.npz` files. The `merge_data` directory contains background–rain composites, while `raw_data` contains background-only events. Each `.npz` stores the arrays `['x', 'y', 't', 'p']`.
|
| 15 |
- **EVK4 recordings (artificial rain & real-world).** Raw `.raw` event streams are converted to `.npz` and segmented into Δt = 0.1 s (10^5 μs per segment), with the same keys `['x', 'y', 't', 'p']`. For the artificial-rain subset, labels are generated using a K-Nearest Neighbors (KNN) procedure: background events in rainy sequences are identified via spatiotemporal alignment with corresponding rain-free data. Label files are provided in `.npy` format.
|
| 16 |
|
|
|
|
| 11 |
|
| 12 |
# EventRain-27K
|
| 13 |
|
| 14 |
+
<p align="center">
|
| 15 |
+
<!-- 横幅：可用相对路径(推荐)或外链 -->
|
| 16 |
+
<img src="assets/EventRain-27K_samples.png" alt="EventRain-27K Overview" width="90%">
|
| 17 |
+
</p>
|
| 18 |
+
|
| 19 |
+
<p align="center">
|
| 20 |
+
<!-- Logo：可放一个或多个 -->
|
| 21 |
+
<a href="https://huggingface.co/datasets/OWNER/REPO">
|
| 22 |
+
<img src="assets/logo.png" alt="EventRain-27K Logo" height="54">
|
| 23 |
+
</a>
|
| 24 |
+
</p>
|
| 25 |
+
|
| 26 |
+
<h1 align="center">EventRain-27K</h1>
|
| 27 |
+
|
| 28 |
+
<p align="center">
|
| 29 |
+
<b>Dataset card for event-based rainy scenes</b>
|
| 30 |
+
</p>
|
| 31 |
+
|
| 32 |
+
---
|
| 33 |
+
|
| 34 |
+
## Source Paper
|
| 35 |
+
|
| 36 |
+
This dataset is derived from the following work:
|
| 37 |
+
|
| 38 |
+
- **PRE-Mamba: A 4D State Space Model for Ultra-High-Frequent Event Camera Deraining**. *ICCV, 2025*. [[paper link]](https://arxiv.org/abs/2505.05307)
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
---
|
| 42 |
+
|
| 43 |
+
## Description
|
| 44 |
+
|
| 45 |
- **Synthetic datasets (KITTI & SPAC).** Event streams are partitioned into contiguous windows of Δt = 0.1 s (10^8 ns per segment) and stored in `.npz` files. The `merge_data` directory contains background–rain composites, while `raw_data` contains background-only events. Each `.npz` stores the arrays `['x', 'y', 't', 'p']`.
|
| 46 |
- **EVK4 recordings (artificial rain & real-world).** Raw `.raw` event streams are converted to `.npz` and segmented into Δt = 0.1 s (10^5 μs per segment), with the same keys `['x', 'y', 't', 'p']`. For the artificial-rain subset, labels are generated using a K-Nearest Neighbors (KNN) procedure: background events in rainy sequences are identified via spatiotemporal alignment with corresponding rain-free data. Label files are provided in `.npy` format.
|
| 47 |
|