[CV] What is Data-Centric approach by LandingAI

大概就是这个意思：

Using image search technology we can find the defect that was missed automatically and find the related similar training images in our training data set that was used to train the model. Once we found those images we can inspect our labels to make sure that they're labeled appropriately. After making changes to the dataset we can retrain our model and observe whether the model is now able to identify this type of defect.
A capability like this requires not just model training and inference but also the ability to search back into a large
unlabeled potentially unlabeled dataset to make the cycle as efficient as possible. Training, inference, search and front-end steps should be as fast as possible. Today there's a long way to go and they're not nearly interactive.