Parametric PerceptNet: A bioinspired deep-net trained for Image Quality Assessment
Jorge Vila-Tomás, Pablo Hernández-Cámara, Valero Laparra, Jesús Malo · 2024
Read Parametric PerceptNet: A bioinspired deep-net trained for Image Quality Assessment by Jorge Vila-Tomás, Pablo Hernández-Cámara, Valero Laparra, Jesús Malo (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-03210.txt.
Description
Jorge Vila-Tomás and co-authors built in 2024 an image quality judge whose layers imitate the human visual system instead of learning freely from scratch. Ordinary deep networks fit human ratings well but carry millions of opaque parameters and overfit. Their parametric network hard-wires bioplausible operations and exposes a small set of meaningful knobs.
The lean models matched the black-box versions on regression against human scores while using about a thousand times fewer parameters. They also trained more smoothly and could be read as vision models, layer by layer. A quirk surfaced in training that the authors name feature spreading, which they take as a warning sign about pure regression training. The paper argues for structure first, data second.
Catalog record
| CREATOR | Jorge Vila-Tomás, Pablo Hernández-Cámara, Valero Laparra, Jesús Malo |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-03210