Electrical Engineering and Systems Science > Image and Video Processing
arXiv:2208.13056 (eess)
[Submitted on 27 Aug 2022 (v1), last revised 25 Mar 2023 (this version, v2)]
Title:Lossy Image Compression with Quantized Hierarchical VAEs
View a PDF of the paper titled Lossy Image Compression with Quantized Hierarchical VAEs, by Zhihao Duan and 3 other authors
View PDFAbstract:Recent research has shown a strong theoretical connection between variational autoencoders (VAEs) and the rate-distortion theory. Motivated by this, we consider the problem of lossy image compression from the perspective of generative modeling. Starting with ResNet VAEs, which are originally designed for data (image) distribution modeling, we redesign their latent variable model using a quantization-aware posterior and prior, enabling easy quantization and entropy coding at test time. Along with improved neural network architecture, we present a powerful and efficient model that outperforms previous methods on natural image lossy compression. Our model compresses images in a coarse-to-fine fashion and supports parallel encoding and decoding, leading to fast execution on GPUs. Code is available atthis https URL.
| Comments: | WACV 2023 Best Algorithms Paper Award, revised version |
| Subjects: | Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) |
| Cite as: | arXiv:2208.13056 [eess.IV] |
| (orarXiv:2208.13056v2 [eess.IV] for this version) | |
| https://doi.org/10.48550/arXiv.2208.13056 arXiv-issued DOI via DataCite | |
| Related DOI: | https://doi.org/10.1109/WACV56688.2023.00028 DOI(s) linking to related resources |
Submission history
From: Zhihao Duan [view email][v1] Sat, 27 Aug 2022 17:15:38 UTC (2,764 KB)
[v2] Sat, 25 Mar 2023 15:52:29 UTC (2,789 KB)
Full-text links:
Access Paper:
- View PDF
- TeX Source
- Other Formats
View a PDF of the paper titled Lossy Image Compression with Quantized Hierarchical VAEs, by Zhihao Duan and 3 other authors
Current browse context:
eess.IV
References & Citations
Bookmark
Bibliographic and Citation Tools
Bibliographic Explorer(What is the Explorer?)
Connected Papers(What is Connected Papers?)
Litmaps(What is Litmaps?)
scite Smart Citations(What are Smart Citations?)
Code, Data and Media Associated with this Article
alphaXiv(What is alphaXiv?)
CatalyzeX Code Finder for Papers(What is CatalyzeX?)
DagsHub(What is DagsHub?)
Gotit.pub(What is GotitPub?)
Hugging Face(What is Huggingface?)
Papers with Code(What is Papers with Code?)
ScienceCast(What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower(What are Influence Flowers?)
CORE Recommender(What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community?Learn more about arXivLabs.