mlsanityData to 7 Oct 2026

Repaint: Improving the Generalization of Down-Stream Visual Tasks by Generating Multiple Instances of Training Examples

arXiv 2110.10366 ↗First posted 20 Oct 2021cs.CV cs.AI cs.LGcited by 5

Not on the map. Why?

Abstract (arXiv)

Convolutional Neural Networks (CNNs) for visual tasks are believed to learn both the low-level textures and high-level object attributes, throughout the network depth. This paper further investigates the `texture bias' in CNNs. To this end, we regenerate multiple instances of training examples from each original image, through a process we call `repainting'. The repainted examples preserve the shape and structure of the regions and objects within the scenes, but diversify their texture and color. Our method can regenerate a same image at different daylight, season, or weather conditions, can have colorization or de-colorization effects, or even bring back some texture information from blacked-out areas. The in-place repaint allows us to further use these repainted examples for improving the generalization of CNNs. Through an extensive set of experiments, we demonstrate the usefulness of the repainted examples in training, for the tasks of image classification (ImageNet) and object detection (COCO), over several state-of-the-art network architectures at different capacities, and across different data availability regimes.

Key facts. Cited by 5 papers in the map (data through 7 Oct 2026). Builds on 28 of its 49 references that are in the map. First posted on arXiv 20 Oct 2021.

Builds on

28 of 49 references.

References · 28

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  3. 2019EfficientNet: Rethinking Model Scaling for Convolutional Neural Networkscited by 6,415
  4. 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
  5. 2015You Only Look Once: Unified, Real-Time Object Detectioncited by 4,829
  6. 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,773
  7. 2016DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFscited by 3,239
  8. 2019CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Featurescited by 2,875
  9. 2017Arbitrary Style Transfer in Real-time with Adaptive Instance Normalizationcited by 1,998
  10. 2017Improved Regularization of Convolutional Neural Networks with Cutoutcited by 1,623
  11. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,430
  12. 2019Semantic Image Synthesis with Spatially-Adaptive Normalizationcited by 1,295
16 more
  1. 2017High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANscited by 1,176
  2. 2016An overview of gradient descent optimization algorithmscited by 1,021
  3. 2019EfficientDet: Scalable and Efficient Object Detectioncited by 1,018
  4. 2015Autoencoding beyond pixels using a learned similarity metriccited by 407
  5. 2018DropBlock: A regularization method for convolutional networkscited by 261
  6. 2019Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNetcited by 260
  7. 2020ClassMix: Segmentation-Based Data Augmentation for Semi-Supervised Learningcited by 209
  8. 2018GAN Augmentation: Augmenting Training Data using Generative Adversarial Networkscited by 107
  9. 2019Implicit Semantic Data Augmentation for Deep Networkscited by 106
  10. 2019Dynamic Instance Normalization for Arbitrary Style Transfercited by 83
  11. 2020Regularizing Deep Networks with Semantic Data Augmentationcited by 73
  12. 2017Cognitive Psychology for Deep Neural Networks: A Shape Bias Case Studycited by 46
  13. 2021Learning to Resize Images for Computer Vision Taskscited by 45
  14. 2020Text and Style Conditioned GAN for Generation of Offline Handwriting Linescited by 35
  15. 2019Light-weight Calibrator: a Separable Component for Unsupervised Domain Adaptationcited by 13
  16. 2020Filter Style Transfer between Photoscited by 10

Built on by

Cited by year

Citing papers per year: 2022: 4; 2023: 1.

5 papers.

  1. 2022FLAME: Free-form Language-based Motion Synthesis & Editingcited by 217
  2. 2022AuxMix: Semi-Supervised Learning with Unconstrained Unlabeled Datacited by 4
  3. 2022Extending Momentum Contrast with Cross Similarity Consistency Regularizationcited by 3
  4. 2022Spatial Cross-Attention Improves Self-Supervised Visual Representation Learningcited by 0
  5. 2023Enhancing Text Generation with Cooperative Trainingcited by 0