Initial Designs

Posted: April 30, 2016 at 4:53 pm

Based on the explorations up to this point, I wanted to post some initial designs using the Gaussian renderer and blending with the original panorama at a more carefully specified horizon. The following images show the results with varying multipliers applied to the Gaussian functions used to draw SOM cells. If you look really closely, you may be able to see that one of the red benches dissolves into a plume of red. A person sitting on one of the red benches on the left turns into thick black smoke.

SOMResults-5_500-tweaked-SOMScale-10_1000000-sigma400-AVG-h1352-scaler0.25 SOMResults-5_500-tweaked-SOMScale-10_1000000-sigma400-AVG-h1352SOMResults-5_500-tweaked-SOMScale-10_1000000-sigma400-AVG-h1352-scaler2

The smaller Gaussians leave empty space in the image (showing white) and allow the viewer to read the width and height of the cells. I prefer the larger Gaussians that fill in these empty spaces. This obliterates some of the structure of the SOM though. The larger Guassians (bottom image) take longer to render, due to much more pixel copying, but I quite like how they break up the rectangular grid of the original panorama and add variation to the edges of the frame.

I plan to write another version of the code that will using only colour features and thus bypass all the segmentation and feature extraction code. I would just down-sample the original image to the max SOM size I can compute on my hardware and seed and train it with the colour values of those pixels. This will be a much simpler process, but the lack of width and height features may simplify the SOM structure; no way to tell without trying. Of course, without a width of height for each cell, the non-rectangular edges would be lost. I’m hoping to do my shoot on Monday and start working on the final image.