突破细粒度分类障碍的渐进式推理阶段数据探索与利用方法

Key Points

• This research aims to propose a new learning paradigm to address barriers in fine-grained classification due to diverse and dynamic data.
• Developed a novel learning paradigm for progressive reasoning during inference.
• Introduced an exploration-utilization strategy named EXP2.
• Investigated the model's ability to utilize low-cost data samples for classification optimization.
• Demonstrated the model's effectiveness in accurately characterizing fine-grained categories.
• Verified the generalizability of the proposed method across various scenarios.
• Showed improvements in classification performance over traditional approaches.