mlsanityData to 7 Oct 2026

Multi-Sample Training for Neural Image Compression

arXiv 2209.13834 ↗First posted 28 Sep 2022cs.CVcited by 5

Not on the map. Why?

Abstract (arXiv)

This paper considers the problem of lossy neural image compression (NIC). Current state-of-the-art (sota) methods adopt uniform posterior to approximate quantization noise, and single-sample pathwise estimator to approximate the gradient of evidence lower bound (ELBO). In this paper, we propose to train NIC with multiple-sample importance weighted autoencoder (IWAE) target, which is tighter than ELBO and converges to log likelihood as sample size increases. First, we identify that the uniform posterior of NIC has special properties, which affect the variance and bias of pathwise and score function estimators of the IWAE target. Moreover, we provide insights on a commonly adopted trick in NIC from gradient variance perspective. Based on those analysis, we further propose multiple-sample NIC (MS-NIC), an enhanced IWAE target for NIC. Experimental results demonstrate that it improves sota NIC methods. Our MS-NIC is plug-and-play, and can be easily extended to other neural compression tasks.

Key facts. Cited by 5 papers in the map (data through 7 Oct 2026). Builds on 41 of its 67 references that are in the map. First posted on arXiv 28 Sep 2022.

Builds on

41 of 67 references.

References · 41

  1. 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 4,993
  2. 2017Neural Discrete Representation Learningcited by 4,650
  3. 2013Auto-Encoding Variational Bayescited by 3,980
  4. 2018Variational image compression with a scale hyperpriorcited by 910
  5. 2016End-to-end Optimized Image Compressioncited by 673
  6. 2018Joint Autoregressive and Hierarchical Priors for Learned Image Compressioncited by 642
  7. 2020Learned Image Compression with Discretized Gaussian Mixture Likelihoods and Attention Modulescited by 570
  8. 2015Importance Weighted Autoencoderscited by 454
  9. 2018DVC: An End-to-end Deep Video Compression Frameworkcited by 324
  10. 2022ELIC: Efficient Learned Image Compression with Unevenly Grouped Space-Channel Contextual Adaptive Codingcited by 296
  11. 2017Lossy Image Compression with Compressive Autoencoderscited by 289
  12. 2019Monte Carlo Gradient Estimation in Machine Learningcited by 281
29 more
  1. 2020Channel-wise Autoregressive Entropy Models for Learned Image Compressioncited by 255
  2. 2021Deep Contextual Video Compressioncited by 216
  3. 2021Checkerboard Context Model for Efficient Learned Image Compressioncited by 193
  4. 2016Variational Lossy Autoencodercited by 175
  5. 2016A Hierarchical Latent Variable Encoder-Decoder Model for Generating Dialoguescited by 174
  6. 2016Full Resolution Image Compression with Recurrent Neural Networkscited by 170
  7. 2015Variable Rate Image Compression with Recurrent Neural Networkscited by 168
  8. 2021FVC: A New Framework towards Deep Video Compression in Feature Spacecited by 149
  9. 2018Inference Suboptimality in Variational Autoencoderscited by 131
  10. 2014Neural Variational Inference and Learning in Belief Networkscited by 116
  11. 2021Enhanced Invertible Encoding for Learned Image Compressioncited by 102
  12. 2017Sticking the Landing: Simple, Lower-Variance Gradient Estimators for Variational Inferencecited by 97
  13. 2019Practical Lossless Compression with Latent Variables using Bits Back Codingcited by 85
  14. 2016Variational inference for Monte Carlo objectivescited by 83
  15. 2020Improving Inference for Neural Image Compressioncited by 81
  16. 2018Tighter Variational Bounds are Not Necessarily Bettercited by 76
  17. 2020Causal Contextual Prediction for Learned Image Compressioncited by 70
  18. 2020Universally Quantized Neural Compressioncited by 60
  19. 2021Soft then Hard: Rethinking the Quantization in Neural Image Compressioncited by 50
  20. 2022Unified Multivariate Gaussian Mixture for Efficient Neural Image Compressioncited by 43
  21. 2018Doubly Reparameterized Gradient Estimators for Monte Carlo Objectivescited by 41
  22. 2020Rao-Blackwellizing the Straight-Through Gumbel-Softmax Gradient Estimatorcited by 37
  23. 2017Reinterpreting Importance-Weighted Autoencoderscited by 36
  24. 2020Compressing Images by Encoding Their Latent Representations with Relative Entropy Codingcited by 36
  25. 2021On the advantages of stochastic encoderscited by 28
  26. 2021Generalized Doubly Reparameterized Gradient Estimatorscited by 14
  27. 2020Infomax Neural Joint Source-Channel Coding via Adversarial Bit Flipcited by 7
  28. 2018Neural Joint Source-Channel Codingcited by 5
  29. 2022Split Hierarchical Variational Compressioncited by 4

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 4; 2025: 0; 2026: 1.

5 papers.

  1. 2024Idempotence and Perceptual Image Compressioncited by 27
  2. 2024Learned Image Transmission with Hierarchical Variational Autoencodercited by 1
  3. 2024Recursive Learning of Asymptotic Variational Objectivescited by 1
  4. 2026TextBoost: Boosting Scene Text Fidelity in Ultra-low Bitrate Image Compressioncited by 1
  5. 2024FLLIC: Functionally Lossless Image Compressioncited by 0