Are we using the correct cell size for our HOG calculation?

[nordstromn508]

One way that we might be able to improve our accuracy is to change the size of our cells. This could improve the HOG's ability to draw out more interesting feature vectors for the Linear SVM to recognize.

[nordstromn508]

Analysis shows that we are currently using the best cell size for our images. Find below, a series of average accuracies with a given cell size averaged over 1000 runs
cell_size: average accuracy
8: 79%
9: 78%
10: 77%
11: 65%
12: 68%
13: 68%
14: 73%
15: 76%
16: 74%