Author: Zane Reynolds (@zreyn) Date: November 2016
Evan's: https://github.com/coradek/CNW_Wildlife_Identification/blob/master/src/feature_extractor.py
- Create the tiles for all of the images, store them in an S3 bucket
- Feature extraction - Inception-v3 from TensorFlow inspired by: http://www.kernix.com/blog/image-classification-with-a-pre-trained-deep-neural-network_p11 ** Read the tiles from S3 ** Define computation graph in TensorFlow ** Run the pre-trained CNN on each tile and save out the next-to-last layer as features
- Define distance metric
- Create distance matrix
- Heirarchical clustering
Install:
- pip install tensorflow
- python src/classify_image.py --model_dir model/imagenet
- add environment variables: AWS_ACCESS_KEY_ID - Your AWS Access Key ID AWS_SECRET_ACCESS_KEY - Your AWS Secret Access Key