ImageNet Classification with Deep Convolutional Neural Networks
Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." Advances in neural information processing systems. 2012.
Models
Using Dataset
Implementation | Accuracy | Weights | Memory | Conv Ops | etc | link |
---|---|---|---|---|---|---|
Keras | 35,659,688 | 35,659,688 * 4bytes | Keras | |||
Tensorflow Slim | 62,378,344 | 62,378,344 * 4bytes | Slim | |||
Pytorch | 90.1% (CIFAR 10, added BN) | 58,323,690 | 58,323,690 * 4bytes | Pytorch |
Tip & Trick
name | for What | reference |
---|---|---|
Relu | faster train / training deeper network | Paper |
LRN | better generalization (replaced with BN later) | LRN in caffe |
Overlapping Pooling | to avoid overfitting | - |
Drop Out | to avoid overfitting | Paper |
Color PCA augmentation | to avoid overfitting, data augmentation | - |
Error of paper
- 224x224 is actually 227x227