LEAD: Learning Decomposition for Source-free Universal Domain Adaptation (2024-03-06T00:00:00.000000Z)

TL;DR

This paper proposes a new idea of LEArning Decomposition (LEAD), which decouples features into source-known and-unknown components to identify target-private data and is complementary to most existing methods.

Abstract

Universal Domain Adaptation (UniDA) targets knowledge transfer in the presence of both covariate and label shifts. Recently, Source-free Universal Domain Adaptation (SF-UniDA) has emerged to achieve UniDA without access to source data, which tends to be more practical due to data protection policies. The main challenge lies in determining whether covariate-shifted samples belong to target-private unknown categories. Existing methods tackle this either through hand-crafted thresholding or by developing time-consuming iterative clustering strategies. In this paper, we propose a new idea of LEArning Decomposition (LEAD), which decouples features into source-known and-unknown components to identify target-private data. Technically, LEAD initially leverages the or-thogonal decomposition analysis for feature decomposition. Then, LEAD builds instance-level decision boundaries to adaptively identify target-private data. Extensive experiments across various UniDA scenarios have demonstrated the effectiveness and superiority of LEAD. Notably, in the OPDA scenario on VisDA dataset, LEAD outperforms GLC by 3.5% overall H-score and reduces 75% time to derive pseudo-labeling decision boundaries. Besides, LEAD is also appealing in that it is complementary to most existing methods. The code is available at https://github.com/ispc-lab/LEAD.

Authors

Florian Rohrbein

3 papers

Changjun Jiang

3 papers

Sanqing Qu

6 papers

TL;DR

Abstract

Authors

References68 items

D2IFLN: Disentangled Domain-Invariant Feature Learning Networks for Domain Generalization

Segment Anything

Upcycling Models Under Domain and Category Shift

Modality-Agnostic Debiasing for Single Domain Generalization

DEJA VU: Continual Model Generalization For Unseen Domains

PSDC: A Prototype-Based Shared-Dummy Classifier Model for Open-Set Domain Adaptation

Divide and Contrast: Source-free Domain Adaptation via Adaptive Contrastive Learning

Unified Optimal Transport Framework for Universal Domain Adaptation

Prior Knowledge Guided Unsupervised Domain Adaptation

POEM: Out-of-Distribution Detection with Posterior Sampling

Geometric Anchor Correspondence Mining with Uncertainty Modeling for Universal Domain Adaptation

Mitigating Neural Network Overconfidence with Logit Normalization

BMD: A General Class-balanced Multicentric Dynamic Prototype Strategy for Source-free Domain Adaptation

A Source-Free Domain Adaptive Polyp Detection Framework With Style Diversification Flow

Training OOD Detectors in their Natural Habitats

UMAD: Universal Model Adaptation under Domain and Category Shift

Exploiting the Intrinsic Neighborhood Structure for Source-free Domain Adaptation

Semantically Coherent Out-of-Distribution Detection

Generalized Source-free Domain Adaptation

Domain Consensus Clustering for Universal Domain Adaptation

OVANet: One-vs-All Network for Universal Domain Adaptation

VDM-DA: Virtual Domain Modeling for Source Data-Free Domain Adaptation

Mining Data Impressions From Deep Models as Substitute for the Unavailable Training Data

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Open-Set Hypothesis Transfer With Semantic Consistency

Learning to Balance Specificity and Invariance for In and Out of Domain Generalization

DACS: Domain Adaptation via Cross-domain Mixed Sampling

Model Adaptation: Unsupervised Domain Adaptation Without Source Data

Universal Source-Free Domain Adaptation

Efficient Domain Generalization via Common-Specific Low-Rank Decomposition

Generalized ODIN: Detecting Out-of-Distribution Image Without Learning From Out-of-Distribution Data

Universal Domain Adaptation through Self Supervision

The EU General Data Protection Regulation (GDPR)

Progressive Domain Adaptation for Object Detection

Universal Domain Adaptation

High-Frequency Component Helps Explain the Generalization of Convolutional Neural Networks

Learning to Transfer Examples for Partial Domain Adaptation

A Review of Domain Adaptation without Target Labels

Moment Matching for Multi-Source Domain Adaptation

ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness

Recognition in Terra Incognita

Domain Invariant and Class Discriminative Feature Learning for Visual Domain Adaptation

Unsupervised Cross-Modality Domain Adaptation of ConvNets for Biomedical Image Segmentations with Adversarial Loss

Open Set Domain Adaptation by Backpropagation

Domain Adaptive Faster R-CNN for Object Detection in the Wild

Maximum Classifier Discrepancy for Unsupervised Domain Adaptation

CyCADA: Cycle-Consistent Adversarial Domain Adaptation

VisDA: The Visual Domain Adaptation Challenge

Open Set Domain Adaptation

Partial Transfer Learning with Selective Adversarial Networks

Deep Hashing Network for Unsupervised Domain Adaptation

No More Discrimination: Cross City Adaptation of Road Scene Segmenters

A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks

The Cityscapes Dataset for Semantic Urban Scene Understanding

Deep Residual Learning for Image Recognition

Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

Domain-Adversarial Training of Neural Networks

Optimal Transport for Domain Adaptation

Transfer Feature Learning with Joint Distribution Adaptation

Adapting Visual Category Models to New Domains

Silhouettes: a graphical aid to the interpretation and validation of cluster analysis

COCA: Classifier-Oriented Calibration for Source-Free Universal Domain Adaptation via Textual Prototype

Supplementary : Learning to Detect Open Classes for Universal Domain Adaptation

Do we really need to access the source data? source hypothesis transfer for un-supervised domain adaptation

A Mathematical Theory of Communication

Decom-pose to adapt: Cross-domain object detection via feature dis-entanglement

license agreement with IEEE. Restrictions apply

Source data-free cross-domain semantic segmentation: Align, teach and propagate

Field of Study

Journal Information

Name

Page

Venue Information

Name

Type

URL