Paper Image

Discovering object parts for fine-grained image recognition

Published on:

6 September 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Robert van der Klis,

Stephan Alaniz,

Massimiliano Mancini,

Cassio F. Dantas,

Dino Ienco,

Zeynep Akata,

Diego Marcos

Bullets

Key Details

PDiscoNet discovers object parts using only image labels, no part annotations

It uses losses to make parts compact, distinct, invariant, and present

The parts form a bottleneck for fine-grained image classification

Results show it finds better parts than previous methods

The parts provide interpretability without hurting classification

AI generated summary

Discovering object parts for fine-grained image recognition

This paper proposes a deep learning method called PDiscoNet that discovers semantically consistent object parts from images, using only image-level labels. It encourages parts to be compact, distinct, invariant to image transforms, and present across some images. This provides interpretability for fine-grained models, without needing part annotations.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up