AP-CNN: Weakly Supervised Attention Pyramid Convolutional Neural Network for Fine-Grained Visual Classification.

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Fine-grained visual classification (FGVC) requires distinguishing subtle differences within object categories.
• Existing FGVC methods primarily leverage high-level features, potentially overlooking crucial low-level details.
• Accurate region localization is critical for effective FGVC performance.

Purpose of the Study:

• To enhance feature representation and region localization in FGVC by integrating low-level visual information.
• To introduce a novel deep learning model, the Attention Pyramid Convolutional Neural Network (AP-CNN), for improved FGVC.
• To demonstrate the efficacy of AP-CNN without requiring additional bounding box or part annotations.

Main Methods:

• Developed a dual pathway hierarchy structure within AP-CNN, combining top-down semantic and bottom-up attention pathways.
• Implemented an ROI-guided refinement strategy, including ROI-guided dropblock and zoom-in operations, to focus on discriminative regions.
• Trained the AP-CNN model end-to-end on standard FGVC datasets.

Main Results:

• The proposed AP-CNN effectively integrates high-level semantic and low-level detailed feature representations.
• ROI-guided refinement successfully enhances discriminative local regions while minimizing background noise.
• Achieved state-of-the-art performance on CUB-200-2011, Stanford Cars, and FGVC-Aircraft datasets.

Conclusions:

• Integrating low-level visual features significantly boosts FGVC performance.
• The AP-CNN architecture provides enhanced feature representation and accurate localization capabilities.
• The model offers a practical solution for FGVC tasks, eliminating the need for complex annotations.