CARE: Constrained Attention Refinement for Fine-Grained Visual Classification via Teacher-Student Distillation

Key points

• Fine-grained visual classification requires models to recognize subtle local traits while exposing the visual evidence behind their predictions.1
• Class-specific attention pathways provide a natural basis for interpretable recognition, but their constrained prediction structure limits discriminative capacity and underuses intermediate representations from strong pretrained backbones.1
• CARE keeps the final prediction and explanation within a class-specific attention student, while introducing a training-only auxiliary query teacher that reads selected intermediate DINOv2 layers with learnable queries.1
• To further refine the explanation pathway, we design diversity and sparsity terms to regularize student attention heads, reducing redundancy and encouraging compact trait localization.1

Sources (1)

• [1]CARE: Constrained Attention Refinement for Fine-Grained Visual Classification via Teacher-Student Distillation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 03:15 AM

  To address this problem, we propose CARE, a constrained attention refinement framework for interpretable fine-grained recognition via teacher-student distillation.

  Fine-grained visual classification requires models to recognize subtle local traits while exposing the visual evidence behind their predictions.

Extractive summary: sentences quoted from the sources.