Unsupervised Learning of Probably Symmetric Deformable 3D Objects From Images in the Wild

We propose a method to learn 3D deformable object categories from raw single-view images, without external supervision. The method is based on an autoencoder that factors each input image into depth, albedo, viewpoint and illumination. In order to disentangle these components without supervision, we use the fact that many object categories have, at least approximately, a symmetric structure. We show that reasoning about illumination allows us to exploit the underlying object symmetry even if the appearance is not symmetric due to shading. Furthermore, we model objects that are probably, but not certainly, symmetric by predicting a symmetry probability map, learned end-to-end with the other components of the model. Our experiments show that this method can recover very accurately the 3D shape of human faces, cat faces and cars from single-view images, without any supervision or a prior shape model. Code and demo available at https://github.com/elliottwu/unsup3d.

Download Full-text

Technology of the Surround

Catalyst Feminism Theory Technoscience ◽

10.28968/cftt.v7i2.35973 ◽

2021 ◽

Vol 7 (2) ◽

Author(s):

Beth Coleman

Keyword(s):

Machine Learning ◽

Machine Vision ◽

Unsupervised Learning ◽

Supervised Learning ◽

Status Quo ◽

Training Data ◽

Reproductive Condition ◽

Game Play ◽

Key Concepts ◽

In The Wild

In addressing the issue of harmful bias in AI systems, this paper asks for a consideration of a generatively wild AI that exceeds the framework of predictive machine learning. The argument places supervised learning with its labeled training data as primarily a form of reproduction of a status quo. Based on this framework, the paper moves through an analysis of two AI modalities—supervised learning (e.g., machine vision) and unsupervised learning (e.g., game play)—to demonstrate the potential of AI as mechanism that creates patterns of association outside of a purely reproductive condition. This analysis is followed by an introduction to the concept of the technology of the surround, where the paper then turns toward theoretical positions that unbind categorical logics, moving toward other possible positionalities—the surround (Harney and Moten), alien intelligence (Parisi), and intra-actions of subject/object resolution (Barad). The paper frames two key concepts in relation to an AI in the wild: the colonial sublime and black techné. The paper concludes with a summation of what AI in the wild can contribute to the subversion of technologies of oppression toward a liberatory potential of AI.

Download Full-text

Guided Unsupervised Learning of Mode Specific Models for Facial Point Detection in the Wild

2014 5th International Conference on Information and Communication Systems (ICICS) ◽

10.1109/iccvw.2013.56 ◽

2013 ◽

Cited By ~ 7

Author(s):

Shashank Jaiswal ◽

Timur R. Almaev ◽

Michel F. Valstar

Keyword(s):

Unsupervised Learning ◽

In The Wild ◽

Point Detection

Download Full-text

Unsupervised learning of 3D objects conserving global topological order

[Proceedings 1992] IEEE International Conference on Systems Engineering ◽

10.1109/icsyse.1992.236951 ◽

2003 ◽

Cited By ~ 1

Author(s):

J. Chao ◽

K. Minowa ◽

S. Tsujii

Keyword(s):

Unsupervised Learning ◽

Topological Order ◽

3D Objects

Download Full-text

Freeze-substituted fungal cells of Nectria haematococca: A comparison of the macroconidial walls of the adhesion-competent wild type with an adhesion-reduced mutant

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s0424820100160996 ◽

1990 ◽

Vol 48 (3) ◽

pp. 688-689

Author(s):

Thecan Caesar-Ton That ◽

Lynn Epstein

Keyword(s):

Freeze Substitution ◽

Uranyl Acetate ◽

Wild Type ◽

Nectria Haematococca ◽

Fruit Extract ◽

Dialysis Tubing ◽

Tube Emergence ◽

Contrast Microscopy ◽

In The Wild ◽

Freeze Damage

Nectria haematococca mating population I (anamorph, Fusarium solani) macroconidia attach to its host (squash) and non-host surfaces prior to germ tube emergence. The macroconidia become adhesive after a brief period of protein synthesis. Recently, Hickman et al. (1989) isolated N. haematococca adhesion-reduced mutants. Using freeze substitution, we compared the development of the macroconidial wall in the wild type in comparison to one of the mutants, LEI.Macroconidia were harvested at 1C, washed by centrifugation, resuspended in a dilute zucchini fruit extract and incubated from 0 - 5 h. During the incubation period, wild type macroconidia attached to uncoated dialysis tubing. Mutant macroconidia did not attach and were collected on poly-L-lysine coated dialysis tubing just prior to freezing. Conidia on the tubing were frozen in liquid propane at 191 - 193C, substituted in acetone with 2% OsO4 and 0.05% uranyl acetate, washed with acetone, and flat-embedded in Epon-Araldite. Using phase contrast microscopy at 1000X, cells without freeze damage were selected, remounted, sectioned and post-stained sequentially with 1% Ba(MnO4)2 2% uranyl acetate and Reynold’s lead citrate. At least 30 cells/treatment were examined.

Download Full-text