DogReID-1553: A Large-Scale Dog Re-Identification Video Dataset

DogReID-1553 is a dataset for dog re-identification across videos and images, containing 7,463 videos of 1,553 distinct dog identities. The dataset was created by Marko Medved for Project Puppies Dataverse and was last updated on 2026-05 06. It includes manually annotated bounding boxes, user-submitted breed labels, and scene-disjoint evaluation splits to prevent data leakage.

Use Cases

• Train or fine-tune dog detection models based on the manually annotated bounding boxes on video first frames.
• Benchmark dog re-identification algorithms based on the provided scene-disjoint query and gallery splits.
• Analyze the relationship between dog breeds and re-identification performance based on the user-submitted breed labels.
• Develop privacy-preserving video processing techniques based on the described anonymization methods.

Strengths

• Contains 7,463 videos across 1,553 distinct dog identities, averaging 4.8 videos per dog.
• Includes manually annotated bounding boxes on the first frame of each video.
• Uses scene-disjoint splits for query and gallery sets to prevent environmental overfitting.
• Videos are released under the permissive Creative Commons CC0 1.0 license.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for annotation files is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the crowdsourced collection method via specific online platforms.

Provenance

Source

Project Puppies Dataverse

Collection Method

Videos were crowdsourced directly from users via a submission website and the Prolific research platform.

Time Range

null

Freshness

Last updated 2026-05-06 18:41:13; freshness should be verified.

Geography

null