FishEye8K: 8,000 Annotated Images for Fisheye Camera Object Detection

8,000 images annotated with 157,000 bounding boxes across five object classes, collected from 18 wide-angle fisheye cameras. The dataset was created by a team of researchers including Munkhjargal Gochoo, Munkh-Erdene Otgonbold, and others. It was last updated on the Hugging Face platform in April 2024.

Use Cases

• Train object detection models based on the 157,000 bounding box annotations.
• Benchmark model performance on fisheye-distorted imagery from the 18 cameras.
• Develop domain adaptation techniques for wide-angle views described in the dataset.
• Research object detection robustness in non-perspective camera geometries.

Strengths

• Contains 8,000 annotated images, providing a substantial base for model training.
• Includes 157,000 bounding box annotations, offering dense object localization data.
• Data originates from 18 different fisheye cameras, suggesting potential diversity in scenes and conditions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's geographic and temporal coverage is unspecified, which may limit understanding of its representativeness.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face user 'hieupth'.

Collection Method

Likely collected from 18 physical fisheye cameras, but the specific gathering method is not detailed.

Time Range

null

Freshness

Last updated 2024-04-08 14:55:14; freshness should be verified.

Geography

null