Cloud4D Dataset

Dataset for "Cloud4D: Estimating Cloud Properties at a High Spatial and Temporal Resolution", NeurIPS 2025 (Spotlight)

Project Page: https://cloud4d.jacob-lin.com/

Dataset Structure

The dataset is organized into two main categories:

Cloud4D/
├── real_world/                    # Real-world stereo camera captures
│   ├── 20230705_10/               # Date-hour folders (YYYYMMDD_HH)
│   │   ├── perspective_1/
│   │   │   ├── left_images/       # Left camera images
│   │   │   ├── right_images/      # Right camera images
│   │   │   ├── camera_pair.npz    # Stereo calibration data
│   │   │   └── *.npy              # Camera intrinsics/extrinsics
│   │   ├── perspective_2/
│   │   └── perspective_3/
│   └── ...
└── synthetic/                     # Synthetic cloud renders
    ├── terragen/                  # Terragen renders
    │   ├── perspective_1/
    │   ├── perspective_2/
    │   └── perspective_3/
    └── large_eddy_simulations/    # LES-based renders
        ├── perspective_1/
        ├── perspective_2/
        ├── perspective_3/
        └── volumes/

Quick Start

Download and Extract

# Clone the dataset
hf download jacoblin/cloud4d --repo-type dataset --local-dir cloud4d
cd cloud4d

# Extract all archives
python unpack.py

# Or extract to a specific location
python unpack.py --output /path/to/cloud4d

Selective Extraction

# Extract only real-world data
python unpack.py --subset real_world

# Extract only synthetic data
python unpack.py --subset synthetic

# Extract a specific date
python unpack.py --filter 20230705

# List available archives without extracting
python unpack.py --list

Parallel Extraction (Faster)

# Use 4 parallel workers
python unpack.py --jobs 4

Citation

If you use this dataset in your research, please cite:

@inproceedings{
      lin2025cloudd,
      title={Cloud4D: Estimating Cloud Properties at a High Spatial and Temporal Resolution},
      author={Jacob Lin and Edward Gryspeerdt and Ronald Clark},
      booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
      year={2025},
      url={https://openreview.net/forum?id=g2AAvmBwkS}