mlsanityData to 7 Oct 2026

On Translation Invariance in CNNs: Convolutional Layers can Exploit Absolute Spatial Location

arXiv 2003.07064 ↗First posted 16 Mar 2020cs.CV cs.LG eess.IVcited by 141

Not on the map. Why?

Abstract (arXiv)

In this paper we challenge the common assumption that convolutional layers in modern CNNs are translation invariant. We show that CNNs can and will exploit the absolute spatial location by learning filters that respond exclusively to particular absolute locations by exploiting image boundary effects. Because modern CNNs filters have a huge receptive field, these boundary effects operate even far from the image boundary, allowing the network to exploit absolute spatial location all over the image. We give a simple solution to remove spatial location encoding which improves translation invariance and thus gives a stronger visual inductive bias which particularly benefits small data sets. We broadly demonstrate these benefits on several architectures and various applications such as image classification, patch matching, and two video classification datasets.

Key facts. Cited by 141 papers in the map (data through 7 Oct 2026). First posted on arXiv 16 Mar 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 44; 2022: 36; 2023: 29; 2024: 11; 2025: 16; 2026: 5.

141 papers.

Sort:

By area:

  1. 2021Alias-Free Generative Adversarial Networkscited by 1,181
  2. 2021TS2Vec: Towards Universal Representation of Time Seriescited by 463
  3. 2021CMT: Convolutional Neural Networks Meet Vision Transformerscited by 310
  4. 2022Long Video Generation with Time-Agnostic VQGAN and Time-Sensitive Transformercited by 244
  5. 2022Scaling Up Your Kernels to 31x31: Revisiting Large Kernel Design in CNNscited by 155
  6. 2021StyleSwin: Transformer-based GAN for High-resolution Image Generationcited by 152
  7. 2021MPViT: Multi-Path Vision Transformer for Dense Predictioncited by 122
  8. 2021FlexConv: Continuous Kernel Convolutions with Differentiable Kernel Sizescited by 66
  9. 2023MLIC++: Linear Complexity Multi-Reference Entropy Modeling for Learned Image Compressioncited by 56
  10. 2021Do Input Gradients Highlight Discriminative Features?cited by 52
  11. 2021Zero-Shot Day-Night Domain Adaptation with a Physics Priorcited by 47
  12. 2023Approximation-Generalization Trade-offs under (Approximate) Group Equivariancecited by 45
38 more
  1. 2022Shift-tolerant Perceptual Similarity Metriccited by 44
  2. 2022Deep Learning Methods for Abstract Visual Reasoning: A Survey on Raven's Progressive Matricescited by 37
  3. 2021Position, Padding and Predictions: A Deeper Look at Position Information in CNNscited by 26
  4. 2022Locality-Aware Inter-and Intra-Video Reconstruction for Self-Supervised Correspondence Learningcited by 26
  5. 2021Segmentation-Renormalized Deep Feature Modulation for Unpaired Image Harmonizationcited by 24
  6. 2022High-fidelity GAN Inversion with Padding Spacecited by 23
  7. 2022On the Strong Correlation Between Model Invariance and Generalizationcited by 23
  8. 2024CSTA: CNN-based Spatiotemporal Attention for Video Summarizationcited by 23
  9. 2022Prior Knowledge Guided Unsupervised Domain Adaptationcited by 20
  10. 2021Convolutional Neural Networks Are Not Invariant to Translation, but They Can Learn to Becited by 19
  11. 2021GANSeg: Learning to Segment by Unsupervised Hierarchical Image Generationcited by 19
  12. 2023Learning to Extrapolate: A Transductive Approachcited by 19
  13. 2023Spatially Covariant Image Registration with Text Promptscited by 18
  14. 2021Toward Spatially Unbiased Generative Modelscited by 17
  15. 2022Leveraging Equivariant Features for Absolute Pose Regressioncited by 17
  16. 2021Shift Invariance Can Reduce Adversarial Robustnesscited by 16
  17. 2021Hallucination In Object Detection -- A Study In Visual Part Verificationcited by 16
  18. 2021Physically Consistent Neural Networks for building thermal modeling: theory and analysiscited by 16
  19. 2021Parametric Scattering Networkscited by 15
  20. 2021Why Do Self-Supervised Models Transfer? Investigating the Impact of Invariance on Downstream Taskscited by 15
  21. 2022The Robustness Limits of SoTA Vision Models to Natural Variationcited by 15
  22. 2023Hyperbolic Contrastive Learningcited by 15
  23. 2021Effects of boundary conditions in fully convolutional networks for learning spatio-temporal dynamicscited by 14
  24. 2022Equivariance and Invariance Inductive Bias for Learning from Insufficient Datacited by 14
  25. 2023What Affects Learned Equivariance in Deep Image Recognition Models?cited by 14
  26. 2025Alias-Free Latent Diffusion Models: Improving Fractional Shift Equivariance of Diffusion Latent Spacecited by 14
  27. 2022Classification of Long Sequential Data using Circular Dilated Convolutional Neural Networkscited by 13
  28. 2022How Tempering Fixes Data Augmentation in Bayesian Neural Networkscited by 13
  29. 2023Evaluating the Robustness of Interpretability Methods through Explanation Invariance and Equivariancecited by 13
  30. 2021Relative stability toward diffeomorphisms indicates performance in deep netscited by 12
  31. 2021Global Pooling, More than Meets the Eye: Position Information is Encoded Channel-Wise in CNNscited by 12
  32. 2022Cylin-Painting: Seamless 360textdegree Panoramic Image Outpainting and Beyondcited by 12
  33. 2024Learning CNN on ViT: A Hybrid Model to Explicitly Class-specific Boundaries for Domain Adaptationcited by 12
  34. 2021Spectral Leakage and Rethinking the Kernel Size in CNNscited by 11
  35. 2021VIPriors 1: Visual Inductive Priors for Data-Efficient Deep Learning Challengescited by 11
  36. 2021Tune It or Don't Use It: Benchmarking Data-Efficient Image Classificationcited by 11
  37. 2021Anti-aliasing Deep Image Classifiers using Novel Depth Adaptive Blurring and Activation Functioncited by 11
  38. 2022Mixed Evidence for Gestalt Grouping in Deep Neural Networkscited by 11

50 most cited of 141 All 141 (JSON)