PhotoBench is the first benchmark constructed from authentic, personal albums, designed to shift the paradigm from visual matching to personalized multi-source intent-driven photo retrieval. This leaderboard provides the complete raw image collection for unrestricted image retrieval research. Unlike the protected version, you are free to choose your own embedding model, caption generator, or agent-based workflow.

Download the dataset ↗ to get started, then submit your prediction results and see how your approach ranks on the leaderboard.