Supervision Interpolation via LossMix: Generalizing Mixup for Object Detection and Beyond
Not on the map. Why?
Abstract (arXiv)
The success of data mixing augmentations in image classification tasks has been well-received. However, these techniques cannot be readily applied to object detection due to challenges such as spatial misalignment, foreground/background distinction, and plurality of instances. To tackle these issues, we first introduce a novel conceptual framework called Supervision Interpolation (SI), which offers a fresh perspective on interpolation-based augmentations by relaxing and generalizing Mixup. Based on SI, we propose LossMix, a simple yet versatile and effective regularization that enhances the performance and robustness of object detectors and more. Our key insight is that we can effectively regularize the training on mixed data by interpolating their loss errors instead of ground truth labels. Empirical results on the PASCAL VOC and MS COCO datasets demonstrate that LossMix can consistently outperform state-of-the-art methods widely adopted for detection. Furthermore, by jointly leveraging LossMix with unsupervised domain adaptation, we successfully improve existing approaches and set a new state of the art for cross-domain object detection.
Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 52 of its 73 references that are in the map. First posted on arXiv 18 Mar 2023.
Builds on
52 of 73 references.
References · 52
- 2015Deep Residual Learning for Image Recognitioncited by 51,809
- 2014Microsoft COCO: Common Objects in Contextcited by 17,498
- 2015Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networkscited by 9,617
- 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
- 2016Feature Pyramid Networks for Object Detectioncited by 5,021
- 2015You Only Look Once: Unified, Real-Time Object Detectioncited by 4,829
- 2015SSD: Single Shot MultiBox Detectorcited by 3,125
- 2015Domain-Adversarial Training of Neural Networkscited by 3,023
- 2019CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Featurescited by 2,875
- 2015Learning Transferable Features with Deep Adaptation Networkscited by 1,323
- 2017CyCADA: Cycle-Consistent Adversarial Domain Adaptationcited by 1,079
- 2021YOLOX: Exceeding YOLO Series in 2021cited by 1,064
38 more
- 2021ByteTrack: Multi-Object Tracking by Associating Every Detection Boxcited by 812
- 2015Return of Frustratingly Easy Domain Adaptationcited by 389
- 2016Unsupervised Domain Adaptation with Residual Transfer Networkscited by 340
- 2021DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentationcited by 299
- 2018Strong-Weak Distribution Alignment for Adaptive Object Detectioncited by 269
- 2020Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixupcited by 260
- 2019Adversarial Domain Adaptation with Domain Mixupcited by 250
- 2018Cross-Domain Weakly-Supervised Object Detection through Progressive Domain Adaptationcited by 218
- 2017Image to Image Translation for Domain Adaptationcited by 171
- 2020Unbiased Mean Teacher for Cross-domain Object Detectioncited by 158
- 2020Seesaw Loss for Long-Tailed Instance Segmentationcited by 146
- 2019Progressive Domain Adaptation for Object Detectioncited by 134
- 2020Exploring Categorical Regularization for Domain Adaptive Object Detectioncited by 130
- 2020Harmonizing Transferability and Discriminability for Adapting Object Detectorscited by 125
- 2021Cross-Domain Adaptive Teacher for Object Detectioncited by 125
- 2020FixBi: Bridging Domain Spaces for Unsupervised Domain Adaptationcited by 124
- 2019Diversify and Match: A Domain Adaptive Representation Learning Paradigm for Object Detectioncited by 122
- 2021Instant-Teaching: An End-to-End Semi-Supervised Object Detection Frameworkcited by 112
- 2021Unsupervised Domain Adaptation of Object Detectors: A Surveycited by 93
- 2022A Closer Look at Smoothness in Domain Adversarial Trainingcited by 85
- 2021AutoMix: Unveiling the Power of Mixup for Stronger Classifierscited by 69
- 2022Safe Self-Refinement for Transformer-based Domain Adaptationcited by 61
- 2020SuperMix: Supervising the Mixing Data Augmentationcited by 58
- 2021DINE: Domain Adaptation from Single and Multiple Black-box Predictorscited by 51
- 2019Bag of Freebies for Training Object Detection Neural Networkscited by 47
- 2020Interpolation-based semi-supervised learning for object detectioncited by 40
- 2021DRANet: Disentangling Representation and Adaptation Networks for Unsupervised Cross-Domain Adaptationcited by 37
- 2019SCL: Towards Accurate Domain Adaptive Object Detection via Gradient Detach Based Stacked Complementary Lossescited by 35
- 2021Unsupervised Domain Generalization by Learning a Bridge Across Domainscited by 33
- 2021PIT: Position-Invariant Transform for Cross-FoV Domain Adaptationcited by 30
- 2021SimROD: A Simple Adaptation Method for Robust Object Detectioncited by 27
- 2022RegMixup: Mixup as a Regularizer Can Surprisingly Improve Accuracy and Out Distribution Robustnesscited by 27
- 2021Both Style and Fog Matter: Cumulative Domain Adaptation for Semantic Foggy Scene Understandingcited by 26
- 2021Leveraging Self-Supervision for Cross-Domain Crowd Countingcited by 24
- 2021Synergizing between Self-Training and Adversarial Learning for Domain Adaptive Object Detectioncited by 24
- 2020Deep Domain Adaptive Object Detection: a Surveycited by 20
- 2021Pareto Domain Adaptationcited by 20
- 2021AFAN: Augmented Feature Alignment Network for Cross-Domain Object Detectioncited by 19
50 most cited of 52 All 52 (JSON)
Built on by
Citing papers per year: 2026: 3.
3 papers.