January 2018
Beginner to intermediate
284 pages
8h 35m
English
For ranking/retrieval-based image generation systems, the ranking position of their original text description can be used for evaluating the system, such as Recall at position k (R@k) and median rank scores, where k = 1,3,5. Recall at position k (R@k) means the percentage of test queries for which a model returns the original item among the top k results. Differently, the median rank represents the k at which a system has a recall of 50% (that is the number of results one would have to consider in order to find the original item for half the queries), therefore, the smaller the k is, the better the performance.
Read now
Unlock full access