Anonymous-ACMMM-2025-Submission/Anonymous_ACMMM_2025_Submission

🗂️ AnonymousACMMM2025_Submission Dataset

This dataset is prepared for the Anonymous ACMMM 2025 submission, containing multi-view event-based data designed for dynamic 3D scene reconstruction tasks.

📁 Dataset Structure

Each subfolder corresponds to a distinct synthetic or real-world scene, such as:

• —lego_6_views/
• —capsule_6_views/
• —garage_6_views/
• —Restroom_6_views/
• —Cubes_6_views/
• —Hinge_6_views/
• —MC-Toy_6_views/
• —Rubik’s-Cube_6_views/

Each scene folder contains 6 views indexed from 1 to 6, and follows this structure:

scene_name_6_views/
├── 1/
│   ├── blurry_images/         # Blurry RGB reference frames (e.g., 0000.png ~ 0039.png)
│   ├── events/                # Event streams in various formats
│   └── train/                 # Additional files for training (if any)
├── 2/
├── ...
├── points3d.ply               # Global sparse point cloud
├── transforms_train.json      # Training camera intrinsics/extrinsics
├── transforms_test.json       # Test camera parameters
├── transforms_train_blurry.json  # Camera poses for blurry image supervision
└── scene.blend                # Optional Blender scene file

📦 Event Data Format

Each sub-view (1/ ~ 6/) contains the following files under events/:

• —event.txt: ASCII-format event stream (timestamp, x, y, polarity)
• —event.h5: Efficient HDF5 version of the event stream
• —event.aedat4: Raw event data in AEDAT4 format (from DAVIS cameras or simulation)

🎨 RGB Image Data

• —blurry_images/: Blurry supervision images, typically aligned with camera poses in transforms_train_blurry.json.

🔧 Applications

This dataset can be used for research in the following areas:

• —Event-based 3D Gaussian Splatting
• —Dynamic scene reconstruction
• —Novel view synthesis from asynchronous data
• —Event-guided video or geometry generation

📚 Loading Examples

Loading `event.txt` (example in Python):

with open("event.txt") as f:
    events = [line.strip().split() for line in f if not line.startswith("#")]

Loading `event.h5`:

import h5py
with h5py.File("event.h5", "r") as f:
    timestamps = f["events"]["t"][:]
    xs = f["events"]["x"][:]
    ys = f["events"]["y"][:]
    polarities = f["events"]["p"][:]

Loading camera poses:

import json
with open("transforms_train.json") as f:
    cam_info = json.load(f)

📄 License

This dataset is released for academic research and peer review only. Redistribution or commercial use is not permitted during the review phase. Please contact the authors for usage after publication.