Machines That See the World More Like Humans Do

 Implications of this?

Home/News/Machines That See the World More Like Humans Do/Full Text

ACM TECHNEWS

Machines That See the World More Like Humans Do

By MIT News, December 14, 2021

The 3D Scene Perception via Probabilistic Programming (3DP3) system developed by Massachusetts Institute of Technology researchers helps machines perceive the world in a more human-like manner.   The probabilistic programming framework enables the system to cross-check detected objects against input data, to determine whether camera-recorded images likely correlate with any candidate scene.

3DP3 can deduce whether mismatches are caused by noise or by errors in scene interpretation that require correction via further processing.  The system represents the scene as a graph, where each object is a node and the lines linking nodes indicate which objects contact one another; the result is more accurate estimations of object arrangement.

3DP3 almost always generated more accurate poses than other deep learning models, and performed significantly better when some objects were partially blocking others.

Full Article. 

Non Line of Sight Technology

Have written about related approaches here a number of times.   NLOS it is called, Non-line-of-Sight technology.   Obvious military applications.   Autonomy data  for complex environments.  And privacy data implications.  With a number of applications around.   Here somewhat new from Stanford.

Scientists Use Lasers to See Inside a Locked Room   By Ryan Whitwam on September 10, 2021 at 7:29 am

The unaided human eye will never be able to see around corners, but anything is possible with enough fancy imaging technology. So-called non-line-of-sight (or NLOS) tech is an increasingly common area of study in the age of self-driving cars, which would benefit hugely from being able to see what’s around the bend. Now, a team from the Stanford Computational Imaging Lab   has taken the idea a step further by spying on objects inside a locked room. All they need is a laser and a keyhole. 

NLOS systems don’t yet have any applications in the real world, but they could eventually help first-responders locate people following a disaster and improve medical imaging. The idea has been validated on numerous occasions, but the current methods are too slow to be useful. In the past, NLOS experiments have relied on flat, reflective surfaces which can bound light around a barrier. The light (usually laser pulses) reflects off the target and back to the experimenters. With some computation, we can get a rough idea of what’s on the other side of the barrier without looking.  .... '