Max Neighbourhood Size and Averaged Top Row

Posted: April 29, 2016 at 10:07 am

After making the decision to seed the top row with averaged values I executed a few runs using a larger network (20px for each SOM cell) and a range of maximum neighbourhood sizes (10, 50, 100, 500, and 1000). The last is so large that it nearly covers the entire width of the map (~1400 cells). Following are a few rectangle renderings of the result. Note, I used too large a scaler, so each cell is drawn quite large. I think this explains the large rectangles at the bottom of the image, but that needs more investigation.

SOMResults-5_500-tweaked-SOMScale-20_1000000-sigma10-AVG SOMResults-5_500-tweaked-SOMScale-20_1000000-sigma50-AVG

SOMResults-5_500-tweaked-SOMScale-20_1000000-sigma100-AVG SOMResults-5_500-tweaked-SOMScale-20_1000000-sigma500-AVG SOMResults-5_500-tweaked-SOMScale-20_1000000-sigma1000-AVG

In the top image, the maximum neighbourhood size (10) was too small to fill in the sky area, despite the inclusion of the seed row at the top. This could be fixed by filling not just the top row, but all cells in the column that don’t contain a segment. Images are dissimilar because each network is trained separately and the max size of the neighbourhood makes significant difference in the structure of the result.

I certainly find the large maximum neighbourhood sizes interesting, but they expand so quickly above the horizon that we lose the gradual transition from readable reality to abstraction. This has inspired two ideas: (a) I like the idea of doing a square image where the readable part of the image is only a small portion at the bottom allowing much more image area for a gradual increase of neighbourhood size above. This is inappropriate for the context of the commision and could be a very large image (8ft x 8ft +). (b) I could preserve the connection with the readable section at the bottom of the image and the abstraction at the top by exponentially, rather than linearly, increasing the neighbourhood size.

With these results I think I’ll be soon ready to do the final shoot as I find myself making a lot of tweaks that could be overly specific to this proxy panorama. I think I have the infrastructure and final approach for final production, right on track.