We reported the test accuracy on ImageNet (ILSVRC2012 Validation Set).
| DataSet | Top-1 | Top-5 | Loss |
|---|---|---|---|
| Both256 | 67.574% | 88.1001% | 1.33896 |
| Shrt256 | 69.0801% | 89.0321% | 1.2711 |
Augmented training and test samples:
This improvement was first described by Andrew Howard [Andrew 2014]. Instead of resizing and cropping the image to 256x256, the image is proportionally resized to 256xN(Nx256) with the short edge to 256. Subcrops of 224x224 are then randomly extracted for training.