Finetuning AlexNet for Oxford-102 Dataset

Disclaimer

This is a work in progress, I am not responsible for any false information provided by the model.

Usage

For Training

Cloning the project

To clone the project, use git clone https://github.com/GhostGengar/AlexNet-Oxford102-Tensorflow.git

The dataset

Download the dataset from here

You will need three files:

• Dataset Images
• The Image Labels
• The data splits

You will also need AlexNet pretrained weights:

• Download from here

Please download the four required files and put them inside the AlexNet folder if you want to train using the .py file.

Start training

Run python train.py inside the AlexNet folder to start training the network.

For Testing

Pretrained Weights

This Tensorflow implementation comes with pretrained weights for Oxford-102, to download the pretrained weights:

• The .data file
• The .index file
• The .meta file

Download all three files and put them inside a new folder named models inside the AlexNet folder.

Pick an image of your choice and put it inside the images folder, rename it to flower.jpg.

Start Testing

To start testing, run python flower_test.py

Requirements

• numpy
• cv2
• scipy
• tensorflow

Explaination

• Full code description is available inside the Oxford_102_AlexNet.ipynb file.

Deployment and Variation

• VGG16 (updating...)
• iOS Deployment (updating...)