CARE Framework Uses DINOv2 Distillation for Fine-Grained Visual Classification
October 9, 2026
CARE employs teacher-student distillation to refine class-specific attention pathways for interpretable image recognition. It uses a training-only auxiliary query teacher that accesses intermediate DINOv2 layers to transfer discriminative knowledge to a student model.
HOW THIS AFFECTS YOU
●
researcherThe method improves discriminative capacity in fine-grained tasks by leveraging multi-level intermediate representations.
●
designerThis approach improves the interpretability of visual recognition models through refined attention pathways.