Open Internet by MindsNet

Continuous Improvement of CNN Models with Unlabeled Data

The author faces a challenge in continuously improving a pre-trained CNN/ResNet model for a computer vision task without access to labeled data. The goal is to leverage incoming unlabeled data to enhance feature representations, prevent model drift, and maintain a production-friendly system.

Computing & Technology, Computer Science, Machine Learning