This is my attempt at classifying images in the cifar10 dataset. It’s based around 3inception modules.
Accuracy
| Architecture | Accuracy on test set |
|---|---|
| c2, c2, mp, c2,c2,mp,f,d,d | xx |
| c2, ap, c2, ap, c2, ap, f, d, d, do | 64.28 %, 60.22 % |
| data-normalized, c2, ap, c2, ap, c2, ap, f, d, d, do | 65.83 % |
| data-normalized-gray, c2, ap, c2, ap, c2, ap, f, d, d, do | 62.91 % |
| inception-module | 62.82 % (stopped becuase of overfitting) |
| inception dense drop | 62.89 % |
| inception inception drop | 74.18 % |
| inceptionx3 dense drop | 77.61 % |
| nceptionx3 dense drop dense drop dense | 77.25 % |
Git
https://github.com/Tzeny/udemy-zero-to-deep-learning/blob/master/course/Cifar10.ipynb