Skip to content

Video Retrieval Evaluation Metrics

1. Top-k Recall (R@k)

Top-k Recall represents the proportion of truly relevant samples that are retrieved within the top k results. In other words, Top-k Recall measures the ratio of relevant samples within the top k search results to the total number of relevant samples. This metric is commonly used to evaluate the accuracy of video retrieval systems, especially in diverse video databases.

Calculation Method

Top-k Recall can be expressed as:

R@k=Number of retrieved relevant samplesTotal number of relevant samples

Here, Number of retrieved relevant samples refers to the count of truly relevant videos among the top k retrieved results, and Total number of relevant samples refers to the total number of relevant videos for the query.

Top-k Recall is an important metric because it evaluates whether the retrieval system can find the most relevant video content within a limited set of results. For example, when k = 1, it indicates whether the system can return a relevant video as the first result; as k increases, the system has more opportunities to return relevant videos among multiple results.