SuperPoint evaluation - Matching Score Metric

[shreyasr-upenn]

Hello,

Thanks for answering my previous questions. I'm trying to build my SuperPoint network from scratch and I want to implement the Matching Score metric. I am finding it hard to understand how the metric is computed. I know you have not implemented it, but if you have understood the metric, would you be able to explain? I'm having a hard time understanding the shared viewpoint region. Is the score the ratio between number of matches vs total number of keypoints detected in the two views?

Could you explain with an example from HPatches?

Thank you!