Zygmunt Pizlo
UC Irvine
| Home |
| CV |
| ViperLab |
| Teaching |
| Center for Theoretical Behavioral Sciences |
| Annual Interdisciplinary Conference |
| ModVis: Computational and Mathematical Modeling in Vision |
|
Professor Zyg's Conjecture: veridical 3D vision is mathematically and computationally so difficult that there is only one way to do it. Once you accept this, it follows that all animals (including us) that see the 3D world veridically (or nearly so) use the same algorithm. A computer that can also see veridically, must be using the algorithm that is used by the human visual system. So, there is no longer any need to wonder about whether computer vision should emulate biological vision. It must. When Gestaltists stated that "the whole is different from the sum of its parts" they meant that the visual system is not linear. Recall that in a linear system, the response to a linear combination of inputs is a linear combination of the responses to the individual inputs. This is not the case in vision. See the demo (courtesy of Prof. Tadamasa Sawada), which shows that the percept of a 2D hexagon and of a 2D "Y junction" cannot explain the percept of a 3D cube. Academic lineage:Zyg's math ancestors. My work is directed by exploring new ideas rather than following established views. In this approach, rational arguments are as important for me as experimental results. The emphasis on principled reasoning means that in my view, cognitive psychology is not a bag of tricks; Neither is my research. A list of my most important contributions is provided in the following file. An abbreviated list is below:1994 - proposed a new theory of shape constancy that is NOT based on "taking slant into account" 1995 - developed a pyramid model explaining the speed-accuracy trade-off in vision and mental size transformation 2000 - developed a pyramid model that shows how human beings solve the Traveling Salesman Problem (TSP) 2001 - published a theoretical paper on inverse problems in vision, making it clear that a priori constraints are at least as important as the information in the retinal image 2008 - published the first coherent treatment of the history of shape perception - 3D Shape book 2009 - introduced a new theory of 3D shape perception based on symmetry, compactness and planarity constraints 2011 - developed a new Bayesian theory of the veridical binocular perception of symmetrical shapes that emphasizes the role played by stereoacuity 2011 - published psychophysical results on the transfer of skilled movement that suggested that the motor system has a pyramidal architecture, very much like the architecture of the visual system 2013 - developed a TSP model with a small human-like working memory 2014 - published a new theory of 3D veridical vision- making a machine that sees like us book 2016 - showed that a perceived closed curve is the shortest path in the log-polar representation (aka complex logarithmic map) present in the primary visual cortex (area V1) of primates 2016 - explained how 3D visual perception can be a "hard science" because symmetry, the least-action principle and the conservation laws operate in 3D vision: book chapter 2016 - helped to explain 3D and 2D figure-ground organization by using 3D symmetry and gravity a priori constraints 2017 - helped to formulate the first fully automated 3D shape recovery model. The model solves the problem by applying symmetry to binocular images: demos 2019 - described a process analogous to Noether's (1918) theorem that explains visual perception (AJP paper) - preprint
You can see the demos for the 2014 book. Check also demos on Tada's web site |
| |
 |
|
|
|
|
