Computer vision: Cosmos — All that is, or was, or ever will be

Computer vision

cosmos 22nd July 2018 at 3:53pm

http://www.movidius.com/

Convolutional neural network, Image processing, Computer graphics

Cloud vision API

Problems

Multi-scale networks and an application.

Scene Parsing with Multiscale Feature Learning, Purity Trees, and Optimal Covers

http://www.robots.ox.ac.uk/~vgg/hzbook/

http://www.clement.farabet.net/research.html#parsing

Hand-eye coordination. See work on grabbing objects in Robotics

http://opencv.org/

Nice notes on mathematical optimization for computer graphics and computer vision

Spiking neural network models

Bio-inspired unsupervised learning of visual features leads to robust invariant object recognition

feedforward feature extraction. studies have suggested that the feedforward information is usually sufficient for invariant object categorization, leading to models like Convolutional neural networks.
biologically plausible learning rules the way that biological visual systems learn the appropriate features has attracted much less attention. Yet the ability of the visual cortex to wire itself, mostly in an unsupervised manner, is remarkable [18] and [19]. Here, we propose that adding bio-inspired learning to bio-inspired architectures could improve the models׳ behavior.

algorithm

The algorithm we used here is a scaled-up version of the one presented in [24] (Unsupervised Learning of Visual Features through Spike Timing Dependent Plasticity). We used a five-layer hierarchical network, with a classifier at the end, similar to HMAX model. _{Specifically, we alternated simple cells that gain selectivity through a sum operation, and complex cells that gain shift and scale invariance through a max operation.} Spike timing -coding (stronger signal fires first). Weight sharing, as in CNNs.

The image is copied and scaled, and presented to copies of the network (like Cortical columns). To increase the sparsity at a given scale and location (corresponding to one cortical column), only the spike corresponding to the best matching orientation is propagated (i.e. a winner-take-all inhibition is employed).