Proxy Anchor-based Unsupervised Learning for Continuous Generalized Category Discovery

Published in

IEEE International Conference on Computer Vision(2023)

External Links:

Generate Graph DownloadPDF

TL;DR

This paper proposes a novel unsupervised class incremental learning approach for discovering novel categories on unlabeled sets without prior knowledge that outperforms the state-of-the-art methods on fine-grained datasets under real-world scenarios.

Abstract

Recent advances in deep learning have significantly improved the performance of various computer vision applications. However, discovering novel categories in an incremental learning scenario remains a challenging problem due to the lack of prior knowledge about the number and nature of new categories. Existing methods for novel category discovery are limited by their reliance on labeled datasets and prior knowledge about the number of novel categories and the proportion of novel samples in the batch. To address the limitations and more accurately reflect real-world scenarios, in this paper, we propose a novel unsupervised class incremental learning approach for discovering novel categories on unlabeled sets without prior knowledge. The proposed method fine-tunes the feature extractor and proxy anchors on labeled sets, then splits samples into old and novel categories and clusters on the unlabeled dataset. Furthermore, the proxy anchors-based exemplar generates representative category vectors to mitigate catastrophic forgetting. Experimental results demonstrate that our proposed approach outperforms the state-of-the-art methods on fine-grained datasets under real-world scenarios.

Authors

Junmo Kim

3 papers

Hyungmin Kim

1 papers

Sungho Suh

1 papers

References55 items

SplitNet: Learnable Clean-Noisy Label Splitting for Learning with Noisy Labels

Grow and Merge: A Unified Framework for Continuous Categories Discovery

XCon: Learning with Experts for Fine-grained Category Discovery

Novel Class Discovery without Forgetting

Class-incremental Novel Class Discovery

Proxy Anchor-based Unsupervised Learning for Continuous Generalized Category Discovery

Published in

IEEE International Conference on Computer Vision(2023)

External Links:

Generate Graph DownloadPDF

TL;DR

Abstract

Authors

Junmo Kim

3 papers

Hyungmin Kim

1 papers

Sungho Suh

1 papers

References55 items

SplitNet: Learnable Clean-Noisy Label Splitting for Learning with Noisy Labels

Grow and Merge: A Unified Framework for Continuous Categories Discovery

XCon: Learning with Experts for Fine-grained Category Discovery

Novel Class Discovery without Forgetting

Class-incremental Novel Class Discovery

Daehwan Kim

1 papers

Hansang Cho

1 papers

DeepDPM: Deep Clustering With an Unknown Number of Clusters

Residual Tuning: Toward Novel Category Discovery Without Labels

Generalized Category Discovery

A Unified Objective for Novel Class Discovery

Novel Visual Category Discovery with Dual Ranking Statistics and Mutual Knowledge Distillation

AutoNovel: Automatically Discovering and Learning Novel Visual Categories

Neighborhood Contrastive Learning for Novel Class Discovery

Joint Representation Learning and Novel Category Discovery on Single- and Multi-modal Data

Rainbow Memory: Continual Learning with a Memory of Diverse Samples

Augmentation Strategies for Learning with Noisy Labels

OpenMix: Reviving Known Knowledge for Discovering Novel Visual Categories in an Open World

ProxyNCA++: Revisiting and Revitalizing Proxy Neighborhood Component Analysis

Proxy Anchor Loss for Deep Metric Learning

DivideMix: Learning with Noisy Labels as Semi-supervised Learning

Automatically Discovering and Learning New Visual Categories with Ranking Statistics

SoftTriple Loss: Deep Metric Learning Without Triplet Sampling

Learning to Discover Novel Visual Categories via Deep Transfer Clustering

Ensemble Deep Manifold Similarity Learning Using Hard Proxies

Understanding and Utilizing Deep Neural Networks Trained with Noisy Labels

Multi-Similarity Loss With General Pair Weighting for Deep Metric Learning

Ranked List Loss for Deep Metric Learning

Multi-class Classification without Multi-class Labels

Learning to cluster in order to Transfer across domains and tasks

No Fuss Distance Metric Learning Using Proxies

Improved Deep Metric Learning with Multi-class N-pair Loss Objective

iCaRL: Incremental Classifier and Representation Learning

Deep Metric Learning via Lifted Structured Feature Embedding

FaceNet: A unified embedding for face recognition and clustering

Learning Fine-Grained Image Similarity with Deep Ranking

Fine-Grained Visual Classification of Aircraft

ImageNet classification with deep convolutional neural networks

Recognizing indoor scenes

Clustering by Passing Messages Between Data Points

Dimensionality Reduction by Learning an Invariant Mapping

Learning a similarity metric discriminatively, with application to face verification

Signature Verification Using A "Siamese" Time Delay Neural Network

The Hungarian method for the assignment problem

Arcface: Additive angular margin for deep face recognition

Novel Dataset for Fine-Grained Image Categorization : Stanford Dogs

Pietro Perona, and Serge Belongie. The caltech-ucsd birds

Learning Multiple Layers of Features from Tiny Images

license agreement with IEEE. Restrictions apply

, To acquire a more clearly separated data, we design a simple network

network comprises a series of the Fully Connected layer (FC) - Batch Normalization (BN) - sigmoid - FC - BN - sigmoid - FC

Add new proxy anchors as the estimated number of novel categories based on c ( · ) results

Assign initial means of newly added proxy anchors based on c ( · ) results

Generate old embedding vectors ˜ z 0 using E 0

Get pseudo-labels D 1 new using a non-parametric clustering approach c ( · ) , affinity propagation, ˆ D 1 new = { ( x 1 new , ˆ y 1 new = argmax( c ( x 1 new )) }

Get pseudo-labels D 1 old using previous network f ˆ D 1 old = { ( x 1 old , ˆ y 1 old = argmax( f 0 ( x 1 old )) }

Get fine separated datasets, D 1 old = { x 1 old } and D 1 new = { x 1 new } on Noisy labeling and Gaussian mixture model using the initial separated dataset B. Fine Split

Field of Study

Computer Science

Journal Information

Name

2023 IEEE/CVF International Conference on Computer Vision (ICCV)

Page

20313-20325

Venue Information

Name

IEEE International Conference on Computer Vision

Type

conference

URL

https://ieeexplore.ieee.org/xpl/conhome/1000149/all-proceedings

Alternate Names

ICCV
IEEE Int Conf Comput Vis
ICCV Workshops
ICCV Work