The code used in the NLI hackathon, 23-24/11/2017 for generating image metadata: face recognition, emotion deteection, age&gender estimation, object detection, year of picutre estimation & textual description
-
Face recognition:
- Extracted facess locations with the HOG algorithm
- Extracted deep features (Using a convolutinal Neural Net) for each face
- Used K-Nearest Neighbors (with supporting ball-tree structure) to find similar faces in our data.
-
Age,Gender&Emotion, Object recogintion: with a CNN trained for each metada field.
-
Textual Description: Using the above meta-data and the nlp library simplenlg.
- Itamar Shenhar [email protected]
- Gil Maman [email protected]
- Dani Rubin [email protected]
- Tom Guter [email protected]