%0 Conference Proceedings %T PARTICUL: Part Identification with Confidence measure using Unsupervised Learning %+ Laboratoire Sûreté des Logiciels (LSL) %+ Modélisation et Recherche d’Information Multimédia [Grenoble] (MRIM ) %+ ScaLable Information Discovery and Exploitation [Grenoble] (SLIDE ) %A Xu-Darme, Romain %A Quénot, Georges %A Chihani, Zakaria %A Rousset, Marie-Christine %Z Accepted at XAIE: 2nd Workshop on Explainable and Ethical AI – ICPR 2022 %< avec comité de lecture %B 2-nd Workshop on Explainable and Ethical AI – ICPR 2022 %C Montréal, Canada %8 2022-08-21 %D 2022 %Z 2206.13304 %K Part detection %K Unsupervised learning %K Interpretability %K Confidence measure %K Fine-grained recognition %Z Computer Science [cs]/Computer Vision and Pattern Recognition [cs.CV] %Z Computer Science [cs]/Machine Learning [cs.LG]Conference papers %X In this paper, we present PARTICUL, a novel algorithm for unsupervised learning of part detectors from datasets used in fine-grained recognition. It exploits the macro-similarities of all images in the training set in order to mine for recurring patterns in the feature space of a pre-trained convolutional neural network. We propose new objective functions enforcing the locality and unicity of the detected parts. Additionally, we embed our detectors with a confidence measure based on correlation scores, allowing the system to estimate the visibility of each part. We apply our method on two public fine-grained datasets (Caltech-UCSD Bird 200 and Stanford Cars) and show that our detectors can consistently highlight parts of the object while providing a good measure of the confidence in their prediction. We also demonstrate that these detectors can be directly used to build part-based fine-grained classifiers that provide a good compromise between the transparency of prototype-based approaches and the performance of non-interpretable methods. %G English %2 https://cea.hal.science/cea-03703962/document %2 https://cea.hal.science/cea-03703962/file/paper.pdf %L cea-03703962 %U https://cea.hal.science/cea-03703962 %~ CEA %~ UGA %~ CNRS %~ INPG %~ LIG %~ OPENAIRE %~ GRID5000 %~ LIG_TDCGE_MRIM %~ DRT %~ CEA-UPSAY %~ LIG-TDCGE-SLIDE %~ UNIV-PARIS-SACLAY %~ LIST %~ MIAI %~ SILECS %~ PNRIA %~ UNIVERSITE-PARIS-SACLAY %~ UGA-EPE %~ ANR %~ GS-ENGINEERING %~ GS-COMPUTER-SCIENCE %~ LIG_SIDCH