flickr27-classifier

Dependencies

• [Python 3] (https://www.python.org/)
• [Tensorflow] (https://www.tensorflow.org/)
• [NodeJS] (https://nodejs.org/en/)

Testing pretrained graph

Run python flickr.py <IMAGE PATH>
For example python flickr.py test/bmw.jpg

Preparations for retraining

Download and extract dataset

• Download the flick27 dataset
• Extract dataset in images/dataset/ directory
• Extraxt images/dataset/flickr_logos_27_dataset_images.tar.gz in images/dataset/flickr_logos_27_dataset_images directory

Preprocess and augment dataset images

• Run npm install in images/preprocess to install node modules
• Run node preprocess_images.js to sort images for retraining
• Run npm install in images/augment to install node modules
• Run node augment.js to augment the dataset images

Retraining

Run python retrain.py
--bottleneck_dir=retrain/bottlenecks
--how_many_training_steps=4000
--model_dir=retrain/inception
--output_graph=retrain/retrained_graph1.pb
--output_labels=retrain/retrained_labels1.txt
--image_dir=images/logos