Random-Access Neural Compression of Material Textures
(2023) In ACM Transactions on Graphics p.1-25- Abstract
- The continuous advancement of photorealism in rendering is accompanied
by a growth in texture data and, consequently, increasing storage and memory demands. To address this issue, we propose a novel neural compression
technique specifically designed for material textures. We unlock two more
levels of detail, i.e., 16× more texels, using low bitrate compression, with
image quality that is better than advanced image compression techniques,
such as AVIF and JPEG XL.
At the same time, our method allows on-demand, real-time decompression
with random access similar to block texture compression on GPUs, enabling
compression on disk and memory. The key idea behind our approach is
compressing multiple material... (More) - The continuous advancement of photorealism in rendering is accompanied
by a growth in texture data and, consequently, increasing storage and memory demands. To address this issue, we propose a novel neural compression
technique specifically designed for material textures. We unlock two more
levels of detail, i.e., 16× more texels, using low bitrate compression, with
image quality that is better than advanced image compression techniques,
such as AVIF and JPEG XL.
At the same time, our method allows on-demand, real-time decompression
with random access similar to block texture compression on GPUs, enabling
compression on disk and memory. The key idea behind our approach is
compressing multiple material textures and their mipmap chains together,
and using a small neural network, that is optimized for each material, to
decompress them. Finally, we use a custom training implementation to
achieve practical compression speeds, whose performance surpasses that of
general frameworks, like PyTorch, by an order of magnitude (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/33eb6ce9-340b-4a31-a9d5-fb73baa016b7
- author
- Vaidyanathan, Karthik ; Salvi, Marco ; Wronski, Bartlomiej ; Akenine-Möller, Tomas LU ; Ebelin, Pontus LU and Lefohn, Aaron
- organization
- publishing date
- 2023-07-26
- type
- Contribution to journal
- publication status
- published
- subject
- in
- ACM Transactions on Graphics
- article number
- 88
- pages
- 25 pages
- publisher
- Association for Computing Machinery (ACM)
- external identifiers
-
- scopus:85166620426
- ISSN
- 0730-0301
- project
- Evaluating and Improving Rendered Visual Experiences
- WASP: Wallenberg AI, Autonomous Systems and Software Program at Lund University
- language
- English
- LU publication?
- yes
- id
- 33eb6ce9-340b-4a31-a9d5-fb73baa016b7
- alternative location
- https://dl.acm.org/doi/pdf/10.1145/3592407
- date added to LUP
- 2023-05-15 08:38:51
- date last changed
- 2024-03-04 15:29:39
@article{33eb6ce9-340b-4a31-a9d5-fb73baa016b7, abstract = {{The continuous advancement of photorealism in rendering is accompanied<br/>by a growth in texture data and, consequently, increasing storage and memory demands. To address this issue, we propose a novel neural compression<br/>technique specifically designed for material textures. We unlock two more<br/>levels of detail, i.e., 16× more texels, using low bitrate compression, with<br/>image quality that is better than advanced image compression techniques,<br/>such as AVIF and JPEG XL.<br/>At the same time, our method allows on-demand, real-time decompression<br/>with random access similar to block texture compression on GPUs, enabling<br/>compression on disk and memory. The key idea behind our approach is<br/>compressing multiple material textures and their mipmap chains together,<br/>and using a small neural network, that is optimized for each material, to<br/>decompress them. Finally, we use a custom training implementation to<br/>achieve practical compression speeds, whose performance surpasses that of<br/>general frameworks, like PyTorch, by an order of magnitude}}, author = {{Vaidyanathan, Karthik and Salvi, Marco and Wronski, Bartlomiej and Akenine-Möller, Tomas and Ebelin, Pontus and Lefohn, Aaron}}, issn = {{0730-0301}}, language = {{eng}}, month = {{07}}, pages = {{1--25}}, publisher = {{Association for Computing Machinery (ACM)}}, series = {{ACM Transactions on Graphics}}, title = {{Random-Access Neural Compression of Material Textures}}, url = {{https://dl.acm.org/doi/pdf/10.1145/3592407}}, year = {{2023}}, }