DP-Net: Learning Discriminative Parts for image recognition

TL;DR

DP-Net is a deep learning architecture that detects discriminative image parts for recognition, combining interpretability with scalability by leveraging pretrained CNNs and a novel part-based module.

Contribution

It introduces a scalable, interpretable part-based recognition system that learns discriminative parts without fine-tuning the CNN, enhancing interpretability and efficiency.

Findings

Effective detection of discriminative parts in images.

Maintains interpretability at image and category levels.

Does not require CNN fine-tuning, improving scalability.

Abstract

This paper presents Discriminative Part Network (DP-Net), a deep architecture with strong interpretation capabilities, which exploits a pretrained Convolutional Neural Network (CNN) combined with a part-based recognition module. This system learns and detects parts in the images that are discriminative among categories, without the need for fine-tuning the CNN, making it more scalable than other part-based models. While part-based approaches naturally offer interpretable representations, we propose explanations at image and category levels and introduce specific constraints on the part learning process to make them more discrimative.