A Forward and Backward Compatible Framework for Few-shot Class-incremental Pill Recognition

TL;DR

The first FSCIPR framework, discriminative and bidirectional compatible few-shot class-incremental learning (DBC-FSCIL), encompasses forward-compatible and backward-compatible learning components and proposes an innovative virtual class generation strategy and a center-triplet (CT) loss to enhance discriminative feature learning.

Abstract

Automatic Pill Recognition (APR) systems are crucial for enhancing hospital efficiency, assisting visually impaired individuals, and preventing cross-infection. However, most existing deep learning-based pill recognition systems can only perform classification on classes with sufficient training data. In practice, the high cost of data annotation and the continuous increase in new pill classes necessitate the development of a few-shot class-incremental pill recognition system. This paper introduces the first few-shot class-incremental pill recognition framework, named Discriminative and Bidirectional Compatible Few-Shot Class-Incremental Learning (DBC-FSCIL). It encompasses forward-compatible and backward-compatible learning components. In forward-compatible learning, we propose an innovative virtual class synthesis strategy and a Center-Triplet (CT) loss to enhance discriminative feature learning. These virtual classes serve as placeholders in the feature space for future class updates, providing diverse semantic knowledge for model training. For backward-compatible learning, we develop a strategy to synthesize reliable pseudo-features of old classes using uncertainty quantification, facilitating Data Replay (DR) and Knowledge Distillation (KD). This approach allows for the flexible synthesis of features and effectively reduces additional storage requirements for samples and models. Additionally, we construct a new pill image dataset for FSCIL and assess various mainstream FSCIL methods, establishing new benchmarks. Our experimental results demonstrate that our framework surpasses existing State-of-the-art (SOTA) methods. The code is available at https://github.com/zhang-jinghua/DBC-FSCIL.

Figures (8)

• Figure 2: The DBC-FSCIL Framework for Pill Recognition: Stage 1 focuses on the virtual class generation and forward-compatible learning; Stage 2 aims at fine-tuning the model for base session classification; Stage 3 is dedicated to the uncertainty-guided synthesis of pseudo old class features for KD, ensuring backward compatibility in the incremental learning process.
• Figure 3: The illustrations of related metric losses. (a) CE loss aims to learn the decision boundary; (b) Triplet loss seeks to constrain the distance between features of the same class to be less than the distance between different classes by a predefined margin; (c) Center loss encourages the intra-class compactness; (d) Our proposed CT loss further promote the intra-class compactness and inter-class separability by directly considering the distance between different class centers.
• Figure 4: Comparison with SOTA methods on FCPill and mCURE. Our method, DBC-FSCIL, comprehensively surpasses other methods.
• Figure 5: Comparison of the confusion matrices of different ablation methods on FCPill and mCURE datasets. FT, FCL, and BCL denote fine-tuning, forward-compatible learning (including VCG and CT loss), and back-compatible learning (including PFS and US).
• Figure 6: Influence of virtual class generation methods. One-fold virtual class generation obtains the best performance.