HoloGAN: Unsupervised Learning of 3D Representations From Natural Images
- Submitting institution
-
The University of Bath
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 203037964
- Type
- E - Conference contribution
- DOI
-
10.1109/ICCV.2019.00768
- Title of conference / published proceedings
- 2019 IEEE/CVF International Conference on Computer Vision (ICCV)
- First page
- 7588
- Volume
- -
- Issue
- -
- ISSN
- 1550-5499
- Open access status
- Exception within 3 months of publication
- Month of publication
- February
- Year of publication
- 2020
- URL
-
-
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
4
- Research group(s)
-
-
- Citation count
- 4
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- HoloGAN is the first generative model that learns 3D representations from natural images in an entirely unsupervised manner. The key contribution is a novel neural network architecture combining a strong inductive bias about the 3D world with deep generative models to learn disentangled representations (pose, shape, and appearance) of 3D objects from images. This work has attracted significant attention from researchers led to an internship of the first author at Adobe Research in the US. This work is published at ICCV, a top-3 computer vision conference ranked 29th by Google Scholar’s Top Publications (2015–2019).
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -