mlsanityData to 7 Oct 2026

Deep Nets with Subsampling Layers Unwittingly Discard Useful Activations at Test-Time

arXiv 2410.01083 ↗First posted 1 Oct 2024cs.CV cs.LGcited by 1

Not on the map. Why?

Abstract (arXiv)

Subsampling layers play a crucial role in deep nets by discarding a portion of an activation map to reduce its spatial dimensions. This encourages the deep net to learn higher-level representations. Contrary to this motivation, we hypothesize that the discarded activations are useful and can be incorporated on the fly to improve models' prediction. To validate our hypothesis, we propose a search and aggregate method to find useful activation maps to be used at test time. We applied our approach to the task of image classification and semantic segmentation. Extensive experiments over nine different architectures on multiple datasets show that our method consistently improves model test-time performance, complementing existing test-time augmentation techniques. Our code is available at https://github.com/ca-joe-yang/discard-in-subsampling.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 49 of its 80 references that are in the map. First posted on arXiv 1 Oct 2024.

Builds on

49 of 80 references.

References · 49

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  3. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
  4. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
  5. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
  6. 2021Swin Transformer: Hierarchical Vision Transformer using Shifted Windowscited by 12,082
  7. 2017Mask R-CNNcited by 7,259
  8. 2016Densely Connected Convolutional Networkscited by 6,691
  9. 2015Rethinking the Inception Architecture for Computer Visioncited by 5,451
  10. 2014Fully Convolutional Networks for Semantic Segmentationcited by 4,845
  11. 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,773
  12. 2020Training data-efficient image transformers & distillation through attentioncited by 4,560
37 more
  1. 2016The Cityscapes Dataset for Semantic Urban Scene Understandingcited by 4,427
  2. 2014Going Deeper with Convolutionscited by 4,231
  3. 2017MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applicationscited by 3,827
  4. 2022A ConvNet for the 2020scited by 3,545
  5. 2018Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentationcited by 3,210
  6. 2016Aggregated Residual Transformations for Deep Neural Networkscited by 2,540
  7. 2017Rethinking Atrous Convolution for Semantic Image Segmentationcited by 2,154
  8. 2019Searching for MobileNetV3cited by 1,984
  9. 2021Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutionscited by 1,960
  10. 2018Unified Perceptual Parsing for Scene Understandingcited by 1,265
  11. 2021Swin Transformer V2: Scaling Up Capacity and Resolutioncited by 1,215
  12. 2017ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devicescited by 1,194
  13. 2018ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Designcited by 1,132
  14. 2016Semantic Understanding of Scenes through the ADE20K Datasetcited by 1,063
  15. 2021PVT v2: Improved Baselines with Pyramid Vision Transformercited by 749
  16. 2023ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoderscited by 703
  17. 2021RepVGG: Making VGG-style ConvNets Great Againcited by 520
  18. 2021MViTv2: Improved Multiscale Vision Transformers for Classification and Detectioncited by 516
  19. 20183D human pose estimation in video with temporal convolutions and semi-supervised trainingcited by 401
  20. 2019Making Convolutional Networks Shift-Invariant Againcited by 390
  21. 2020BatchEnsemble: An Alternative Approach to Efficient Ensemble and Lifelong Learningcited by 368
  22. 2017Snapshot Ensembles: Train 1, get M for freecited by 349
  23. 2021XCiT: Cross-Covariance Image Transformerscited by 335
  24. 2022EfficientFormer: Vision Transformers at MobileNet Speedcited by 240
  25. 2021Co-Scale Conv-Attentional Image Transformerscited by 224
  26. 2021VOLO: Vision Outlooker for Visual Recognitioncited by 164
  27. 2023Diverse Data Augmentation with Diffusions for Effective Test-time Prompt Tuningcited by 150
  28. 2020Better Aggregation in Test-Time Augmentationcited by 121
  29. 2020Greedy Policy Search: A Simple Baseline for Learnable Test-Time Augmentationcited by 61
  30. 2020Learning Loss for Test-Time Augmentationcited by 56
  31. 2013Stochastic Pooling for Regularization of Deep Convolutional Neural Networkscited by 48
  32. 2023TeSLA: Test-Time Self-Learning With Automatic Adversarial Augmentationcited by 28
  33. 2021Diversity Matters When Learning From Ensemblescited by 27
  34. 2022Learnable Polyphase Sampling for Shift Invariant and Equivariant Convolutional Networkscited by 18
  35. 2020Classification Confidence Estimation with Test-Time Data-Augmentationcited by 11
  36. 2023TAAL: Test-time Augmentation for Active Learning in Medical Image Segmentationcited by 10
  37. 2023Joint Optimization of Class-Specific Training- and Test-Time Data Augmentation in Segmentationcited by 3

Built on by

Cited by year

Citing papers per year: 2025: 1.

1 paper.

  1. 2025Group Downsampling with Equivariant Anti-aliasingcited by 3