AlexNet model from ILSVRC 2012
This repository comes with AlexNet's implementation in TensorFlow. AlexNet is the winner of the ILSVRC-2012 Competition.
The original model introduced in the paper used two separate GPUs for architecturing. That was due to the lack of GPU resources (memory) at the time. Because the limitation is no longer applicable for the current GPU technology for the AlexNet, this repository's implementation merged two separate models allocated into two separate GPUs into one.
python alexnet.py
import cifar10_utils
from alexnet import AlexNet
...
valid_set = (valid_features, valid_labels)
...
alexNet = AlexNet('cifar10', learning_rate=0.0001)
alexNet.train(epochs=20,
batch_size=128,
valid_set=valid_set,
save_model_path='./model')
Environment
Approximate running time
Hyperparameters
Test Accuracy: 0.6548566878980892
1. Input Layer of Image Size (224 x 224 x 3)
2. Convolutional Layer (96 x (11 x 11 x 3)) + stride size of 4
3. Convolutional Layer (256 x (5 x 5 x 48))
4. Convolutional Layer (384 x (3 x 3 x 128))
5. Convolutional Layer (384 x (3 x 3 x 192))
6. Convolutional Layer (256 x (3 x 3 x 192))
7. Fully Connected Layer (4096)
8. Fully Connected Layer (4096)
9. Fully Connected Layer (1000)