Exploration Graph for Multimedia Retrieval

At ICMR’24, we introduced the Continuous Refinement Exploration Graph (crEG), a novel approach designed for efficient multimedia retrieval. Despite being an approximate nearest neighbor search algorithm, crEG is optimized for achieving high recall rates in large-scale multimedia datasets. It offers retrieval precision of 99% while being orders of magnitude faster than linear search methods.

 

Contribution

• Defines the structure of an Exploration Graph.
• Presents an algorithm for incremental graph extension.
• Introduces a method to continuously refine an existing exploration graph, enhancing retrieval quality.
• Scales efficiently to handle large multimedia datasets without sacrificing speed.
• Reduces computational overhead compared to traditional retrieval techniques.

 

Results

In our experiments, crEG outperformed state-of-the-art methods in search and exploration tasks by up to 250% on complex datasets, demonstrating its robustness in real-world multimedia retrieval scenarios. Additionally, crEG's construction time is 2-3 times faster than competing approaches.

More information can be found in our Paper.