COCO2017: Object Detection Dataset with 80 Classes

benjamintli uploaded a dataset containing object detection annotations converted to the HuggingFace image dataset format. The dataset includes 80 classes such as person, bicycle, car, and motorcycle, and is split into training and validation sets. All credits for the original data go to the COCO2017 dataset creators.

Use Cases

• Train object detection models based on the 80 annotated classes.
• Benchmark model performance on a standard validation split.
• Fine-tune pre-trained vision models on a diverse set of common objects.

Strengths

• Contains annotations for 80 distinct object classes.
• Provides a standard split into training and validation subsets.

Limitations

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

Provenance

Source

Original COCO2017 dataset creators.

Collection Method

Converted to HuggingFace image dataset format via hubify.

Freshness

Last updated 2026-01-05 04:46:33; freshness should be verified.