Online experiment

In our previous 2 experiments we confirmed Enhanced MoCo is the best choice for this project. Here we take a look at the actual improvement after our official launch.
Note that, in item matching project, racall and precision are normally a trade-off. It's very hard to improve both at the same time. But if that happens, it's a great success!

Enhanced MoCo's improvement againt status-quo

There are 2 metrics that we care about the most for the actual downstream applications. The total amount of identical items that we could capture through vision: reflected by Coverage. Among these recalled items, how many of them are true positive matches: reflected by Precision.

  • Enhanced MoCo is ablt to improve both coverage and precision which excatly what we are looking for.
  • We don't need any manpower for dataset labeling as this is all done in self-supervised manner.
  • GPUs training is still considerably long. There is definetly room to improve the algorithm.