Oscar-Huang/CarlaCORL_Parquet

Dataset Structure:

CarlaCORL/
├── highway/
│   └── weather-0/
│       ├── data/              # standard routes
│       │   └── routes_xxx/
│       │       ├── actors_data/
│       │       ├── measurements/
│       │       ├── rgb_front/
│       │       └── ...
│       ├── failed/            # failed routes (if any)
│       ├── under_perform/     # underperforming routes (if any)
│       └── validation/        # validation routes (if any)
├── rural/
└── urban/

Load with Hugging Face Datasets

from datasets import load_dataset

# Load entire dataset
dataset = load_dataset("OscarHuang/CarlaCORL")

# Stream loading (without downloading all data)
dataset = load_dataset("OscarHuang/CarlaCORL", streaming=True)

# Access data
for sample in dataset['train']:
    domain = sample['domain']
    frame_id = sample['frame_id']

    # Images are automatically decoded
    rgb_front = sample['rgb_front']  # PIL.Image or bytes

    # JSON data needs parsing
    import json
    measurements = json.loads(sample['measurements'])

Extract as Datasets (RGB, Bev, Measurements...)

python restore_from_parquet.py
--input_dir /your directory/CarlaCORL
--output_dir /your directory/CarlaCORL_Full \