Update README.md

README.md

@@ -1,3 +1,254 @@
+---
+license: cc-by-nc-nd-4.0
+tags:
+- >-
+  - event-camera   - event-based-vision   - synthetic-data   - path-tracing   -
+  neuromorphic-vision   - computer-vision   - event-to-video   -
+  event-to-depth   - sim2real
 ---
-license: cc-by-nc-nd-4.0
----
+
+# ETScenes: Event-RGB Dataset for Fast Path Tracing-based Event Stream Rendering
+
+[![arXiv](https://img.shields.io/badge/arXiv-2508.18071-b31b1b.svg)](https://arxiv.org/abs/2508.18071)
+[![Dataset](https://img.shields.io/badge/HuggingFace-ETScenes-yellow)](https://huggingface.co/datasets/andrewbxy/ETScenes)
+[![License: CC BY-NC-ND 4.0](https://img.shields.io/badge/License-CC--BY--NC--ND--4.0-lightgrey.svg)](https://creativecommons.org/licenses/by-nc-nd/4.0/)
+
+**ETScenes** is a synthetic event-RGB dataset released with our paper:
+
+> **EventTracer: Fast Path Tracing-based Event Stream Rendering**  
+> Zhenyang Li*, Xiaoyang Bai*, Jinfan Lu, Pengfei Shen, Edmund Y. Lam, Yifan Peng  
+> arXiv 2025
+
+ETScenes is generated by **EventTracer**, a path tracing-based event rendering pipeline designed to produce high-fidelity event streams from complex 3D scenes. Unlike conventional video-to-event simulators that rely on low-frame-rate RGB videos, EventTracer directly renders temporally dense event data from 3D scenes using low-sample-per-pixel path tracing and a lightweight physics-aware spiking network.
+
+This dataset is intended to support research in:
+
+- Event-based vision
+- Synthetic event data generation
+- Event-to-video reconstruction
+- Event-based depth estimation
+- Real2Sim / Sim2Real evaluation
+- Neuromorphic vision under complex lighting and high-frequency motion
+- Physically grounded rendering for event cameras
+
+---
+
+## Why ETScenes?
+
+Event cameras asynchronously capture per-pixel brightness changes with extremely high temporal resolution and high dynamic range. However, collecting large-scale paired event-RGB data with dense annotations is expensive, hardware-dependent, and difficult to scale.
+
+ETScenes addresses this challenge by providing synthetic event-RGB data rendered from 3D scenes, where the rendering pipeline naturally supports controlled camera trajectories, complex illumination, and extensible annotations.
+
+Compared with conventional video-to-event pipelines, ETScenes is designed to better preserve:
+
+- Fine structural details
+- High-frequency visual patterns
+- Fast temporal changes
+- Complex lighting effects
+- Event distributions closer to real-world event data
+
+In our paper, ETScenes is used to evaluate the fidelity of simulated event streams through downstream **Real2Sim** tests on event-to-video reconstruction and event-based depth estimation.
+
+---
+
+## Dataset Highlights
+
+- **Path-traced synthetic scenes** with realistic global illumination.
+- **Paired RGB / EXR / event data** for event-based vision research.
+- **Multiple event simulation variants**, including EventTracer outputs and baseline simulator outputs.
+- **EvSNet training data** for studying event stream generation from low-SPP path-traced videos.
+- **Scene diversity**, including indoor, outdoor, day, night, fast-motion, kitchen, classroom, and staircase scenes.
+- **Large-scale release**, approximately 105GB in total.
+
+---
+
+## Repository Structure
+
+The dataset is organized by scene folders. A typical scene folder contains rendered images, HDR EXR images, and event streams generated by different simulation methods.
+
+```text
+ETScenes/
+├── Bistro-Exterior-Day/
+│   ├── images/
+│   ├── images_exr/
+│   ├── events-ours.npz
+│   ├── events-v2e.npz
+│   ├── events-v2ce.npz
+│   ├── events-optix.npz
+│   ├── events-nrd.npz
+│   └── events-dvm.npz
+│
+├── Bistro-Exterior-Dark/
+├── Bistro-Exterior-Fast/
+├── Bistro-Exterior-Night/
+├── Bistro-Interior/
+├── classroom/
+├── kitchen/
+├── staircase/
+│
+└── EvSNet-training/
+    ├── Bistro-Exterior-Day_2048SPP_events/
+    ├── Bistro-Exterior-Day_64SPP_images/
+    ├── Bistro-Exterior-Night_2048SPP_events/
+    ├── Bistro-Exterior-Night_64SPP_images/
+    ├── Bistro-Interior_2048SPP_events/
+    ├── Bistro-Interior_64SPP_images/
+    ├── classroom_2048SPP_events/
+    ├── classroom_64SPP_images/
+    ├── kitchen_2048SPP_events/
+    ├── kitchen_64SPP_images/
+    ├── staircase_2048SPP_events/
+    └── staircase_64SPP_images/
+```
+
+### File naming
+
+| File / Folder | Description |
+|---|---|
+| `images/` | Standard rendered RGB frames. |
+| `images_exr/` | High-dynamic-range EXR rendered frames. |
+| `events-ours.npz` | Event streams generated by EventTracer. |
+| `events-v2e.npz` | Event streams generated by V2E baseline. |
+| `events-v2ce.npz` | Event streams generated by V2CE baseline. |
+| `events-optix.npz`, `events-nrd.npz`, `events-dvm.npz` | Additional rendering / simulation variants used for comparison or ablation. |
+| `EvSNet-training/` | Training data for the lightweight event spiking network used in EventTracer. |
+
+---
+
+## Download
+
+### Install dependencies
+
+```bash
+pip install -U huggingface_hub hf_xet
+```
+
+### Download the full dataset
+
+```bash
+hf download andrewbxy/ETScenes \
+  --repo-type dataset \
+  --local-dir ETScenes
+```
+
+### Download a single scene
+
+For example, to download only `Bistro-Exterior-Day`:
+
+```bash
+hf download andrewbxy/ETScenes \
+  --repo-type dataset \
+  --include "Bistro-Exterior-Day/*" \
+  --local-dir ETScenes
+```
+
+To download the EvSNet training subset:
+
+```bash
+hf download andrewbxy/ETScenes \
+  --repo-type dataset \
+  --include "EvSNet-training/*" \
+  --local-dir ETScenes
+```
+
+---
+
+## Quick Start
+
+The event files are stored as `.npz` files. You can inspect the available keys as follows:
+
+```python
+import numpy as np
+
+path = "ETScenes/Bistro-Exterior-Day/events-ours.npz"
+
+data = np.load(path, allow_pickle=True)
+print(data.files)
+
+for key in data.files:
+    arr = data[key]
+    print(key, type(arr), getattr(arr, "shape", None), getattr(arr, "dtype", None))
+```
+
+Since different downstream pipelines may expect different event representations, we recommend first inspecting the `.npz` keys and then adapting the loading script to your model or benchmark format.
+
+---
+
+## Recommended Use Cases
+
+ETScenes can be used for:
+
+1. **Benchmarking event simulators**  
+   Compare EventTracer against V2E, V2CE, and other video-to-event or scene-to-event simulation methods.
+
+2. **Event-to-video reconstruction**  
+   Evaluate whether pretrained event-to-video models can reconstruct high-quality intensity frames from synthetic event streams.
+
+3. **Event-based depth estimation**  
+   Study how realistic synthetic events transfer to geometric perception tasks.
+
+4. **Real2Sim fidelity evaluation**  
+   Test how closely synthetic event data matches real-world event distributions by evaluating real-data-pretrained models on simulated data.
+
+5. **Training and validating event generation models**  
+   Use the EvSNet training subset to study event stream generation from noisy low-SPP path-traced images.
+
+6. **Rendering-aware neuromorphic vision**  
+   Explore how global illumination, HDR lighting, and high-frequency textures affect event camera simulation.
+
+---
+
+## Relation to EventTracer
+
+ETScenes is released as part of the EventTracer project. EventTracer is a fast path tracing-based event stream rendering pipeline. It combines:
+
+- Low-SPP Monte Carlo path tracing for efficient temporally dense rendering.
+- A lightweight event spiking network, **EvSNet**, for converting noisy path-traced pixel illuminance into event streams.
+- A bipolar leaky integrate-and-fire unit, **BiLIF**, to better model event sensor behavior.
+- A bidirectional earth mover’s distance loss to train event sequence generation.
+
+The goal is to generate synthetic event streams that are both efficient to render and more faithful to real event camera outputs.
+
+---
+
+## Dataset Scope and Limitations
+
+ETScenes is primarily designed for research on synthetic event generation and event-based vision benchmarking. It is not a replacement for all real-world event camera datasets.
+
+Please note:
+
+- The dataset is synthetic and rendered from 3D scenes.
+- The current release focuses on scene-level event-RGB data and simulator comparisons.
+- Large-scale Sim2Real training may require additional scene diversity, more dynamic objects, and broader real-world validation.
+- Users should carefully evaluate domain gaps before deploying models trained only on ETScenes to real sensors.
+
+---
+
+## License
+
+This dataset is released under the **Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License**.
+
+You may use this dataset for academic and non-commercial research purposes under the terms of the license. Please check the license carefully before redistribution or commercial use.
+
+---
+
+## Citation
+
+If you find ETScenes or EventTracer useful for your research, please cite our paper:
+
+```bibtex
+@article{li2025eventtracer,
+  title={EventTracer: Fast Path Tracing-based Event Stream Rendering},
+  author={Li, Zhenyang and Bai, Xiaoyang and Lu, Jinfan and Shen, Pengfei and Lam, Edmund Y. and Peng, Yifan},
+  journal={arXiv preprint arXiv:2508.18071},
+  year={2025}
+}
+```
+
+---
+
+## Contact
+
+For questions, issues, or discussions, please open a discussion on this Hugging Face dataset page.
+
+Project page / code release will be updated here when available.