Alexnet with Tensorflow.

This project is based on Alexnet.
Tensorflow is used to build alexnet network and used Pillow and numpy to load and convert dataset.

https://www.nvidia.cn/content/tesla/pdf/machine-learning/imagenet-classification-with-deep-convolutional-nn.pdf

Requirements

This project used pipenv so you can check pipfile in repository.
you can just install all requirements with pipenv install.

• numpy: 1.14.5
• pandas: 0.23.1
• pillow: 5.1.0
• tensorflow: 1.8.0

Dataset

For training alexnet flower dataset used. There are 4242 images about 5 classes of flowers.
It's divided in directory of flower names so easy to handle it.

• https://www.kaggle.com/alxmamaev/flowers-recognition

Training

You can train model with train.py. Before execute train you have to prepare dataset. So download it into resources then unzip it. training_dataset.csv and valid_dataset, etc are splitted dataset with ratio. train.py train model with above csv file.

• you can change dropout rate and when model is creating.
• tensorboard log will be saved in logdir
• weights will be saved in weights

Inference

It doesn't have command yet. so modify image_path in inference.py then execute it.

• first output: [[ -9.98178387 -14.03773689 8.21770668 -0.54801857 -3.23122072]]
• index of class: 2
• The class of image is rose