Wide Residual Networks in Keras
This is a tag for the final commit to the old version of Wide Residual Networks. This was the implementation which deviates from the paper in the order of concatenations, and therefore performance of these models is a little less than the paper.
The checkpoint is only for old weights purposes. I would not recommend using the older models, even if weights are available for them.
Weights for the WRN-28-8 model.
Lightly improved accuracy (95.54 % compared earlier models 95.08%)