Fig. 1: Overview of the training phase in our network framework.
From: Material classification method of traditional Chinese painting image based on prototypical network
The process consists of support and query set creation, data enhancement via cropping (for the material task) and zooming (for the auxiliary task), ResNet18-based feature extraction, prototype representation calculation, and final classification through a voting mechanism. The inference phase follows the same pipeline without auxiliary supervision, where the final prediction is based on the distance between query samples and learned prototypes.
