Improving filling level classification with adversarial training
This work provides an incremental improvement in filling level classification for robotic manipulation and inventory management systems, particularly for transparent containers.
This paper addresses the challenge of classifying the filling level of cups or drinking glasses from single images, a task complicated by transparency, shape variations, occlusions, and limited training data. The authors propose a transfer learning strategy using adversarial training on a generic source dataset, followed by refinement with a task-specific dataset, which consistently improved classification accuracy and reduced overfitting.
We investigate the problem of classifying - from a single image - the level of content in a cup or a drinking glass. This problem is made challenging by several ambiguities caused by transparencies, shape variations and partial occlusions, and by the availability of only small training datasets. In this paper, we tackle this problem with an appropriate strategy for transfer learning. Specifically, we use adversarial training in a generic source dataset and then refine the training with a task-specific dataset. We also discuss and experimentally evaluate several training strategies and their combination on a range of container types of the CORSMAL Containers Manipulation dataset. We show that transfer learning with adversarial training in the source domain consistently improves the classification accuracy on the test set and limits the overfitting of the classifier to specific features of the training data.