Classic Paper Day #2: Steerable Filter Double Header + a special bonus paper

 The much neglected classic paper day tag resurfaces for a double feature presentation. Papers anyone working with convolutional neural networks (especially visualizing their feature maps) should probably be very familiar with (but you get the sense that is probably not true for many of the younger people).

We will start with the classic paper by Freeman and Adelson titled 'The Design and Use of Steerable Filters'. From a PAMI Transactions issue way back in 1991.  You can find the pdf here.

And of course no read through is complete without David Heeger's notes on Steerable Filters from almost a decade later.

And we immediately segue into the Steeraable Pyramid.  And the followup classic paper titled 'The Steerable Pyramid: A Flexible Architecture for Multi-Scale Derivative Computation' by Simoncelli and Freeman. You can find the pdf here.

Eero Simoncelli has a great set of notes on steerable pyramids you can get here.

It's probably worth also pointing out another paper that came out around the same time period as that lat one titled 'Pyramid-Based texture Analysis/Synthesis' by Heeger and Bergen.  You can find that one here.

Now i wonder why this material should be essential background reading for anyone who has a hankering to learn about CNNs? I wonder...


Popular posts from this blog

Simulating the Universe with Machine Learning

CycleGAN: a GAN architecture for learning unpaired image to image transformations

Pix2Pix: a GAN architecture for image to image transformation