Feature classification service

Service extracts features from images and stores it in feature dataset for classification purposes.

Dependencies

• python 3.9
• tensorflow 2.10.0
• tensorflow-hub 0.12.0
• flask 2.2.2
• numpy 1.23.3
• opencv-python 4.6.0.66
• requests 2.28.1

Deploy

1. Execute run flask in project directory
2. For running scripts that represent service functionality, execute python tests/*.py

Feature extractor

Model is taken from tensorflow hub, url is set in the settings. For now, model is efficient-net-lite0, output dimension is 1280.

Feature classification

Suppose we are given an input vector V that we want to classify. For every feature vector U from dataset:

1. Calculates difference (U-V)
2. Calculates squared euclidean norm of the difference

Label with minimal norm is chosen as response.