Surround-View Parking Dataset with 40K Multi-Camera Frames

ParkRecon3D is a surround-view dataset for parking-aware reconstruction. It contains 40,000 synchronized multi-camera sensor frames and 60,000 human-annotated parking-slot labels from four fisheye cameras. The data was collected in four representative underground parking scenes.

Use Cases

• Train a 3D reconstruction model using synchronized image streams from four fisheye cameras.
• Develop a parking-slot detection system using the 60,000 human-annotated parking-slot labels.
• Build a surround-view perception model for autonomous driving using the 360° coverage of underground parking scenes.
• Analyze parking-slot layout patterns across the four representative underground parking scenes included in the dataset.

Strengths

• Contains 40,000 multi-camera sensor frames, providing substantial data for model training.
• Includes 60,000 human-annotated parking-slot labels for supervised learning tasks.
• Covers 360° surroundings via synchronized streams from four fisheye cameras.
• Represents four distinct underground parking scenes, offering environmental variety.

Limitations

• Dataset scope is limited to four specific underground parking scenes, which may not generalize to all parking environments.
• The raw description does not specify the resolution, frame rate, or calibration details for the fisheye camera streams.
• Unknown file formats and data size may complicate initial integration and processing.

Provenance

Source

Author pipa83165 on Hugging Face.

Collection Method

Synchronized image streams collected from four fisheye cameras in underground parking scenes; labels are human-annotated.

Freshness

Last updated on 2026-03-10, indicating recent maintenance.

Geography

Data collected from underground parking scenes; specific geographic location is unknown.