Human Person Detection on COCO 2017 with Pre-Trained Model Weights

Filtered COCO 2017 person split provides a specialized subset for human detection tasks. The dataset includes trained model weights for YOLOv8, Faster R-CNN, and HOG+SVM architectures. It is hosted on Kaggle, but details on the original author, license, and dataset size are not provided.

Use Cases

• Benchmarking object detection models based on the provided YOLOv8, Faster R-CNN, and HOG+SVM weights.
• Training specialized person detectors based on the filtered COCO 2017 annotations.
• Comparing traditional (HOG+SVM) and deep learning (YOLOv8, Faster R-CNN) approaches for the same detection task.
• Fine-tuning pre-trained models for downstream applications requiring human localization.

Strengths

• Includes trained weights for three distinct detection architectures: YOLOv8, Faster R-CNN, and HOG+SVM.
• Focuses on a filtered 'person' split from the widely recognized COCO 2017 dataset.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Kaggle, derived from the COCO 2017 dataset.

Collection Method

Filtered subset of the COCO 2017 annotations, likely focusing on the 'person' category.

Time Range

Based on COCO 2017, the underlying imagery likely dates to around 2017.

Freshness

Last update date is unknown; freshness unverified.

Geography

The COCO dataset contains images from diverse global contexts; specific geography for this filtered split is unknown.