Echoliu27 / Ai_bias

This repo a collections of repo:

Ai_bias

This repo explores a collections of repos about emotion recognition and ethnicity classifer.

The two experimentations folder explores two public repos:

1. Facial-Expression-Recognition.Pytorch: https://github.com/WuJie1010/Facial-Expression-Recognition.Pytorch
2. Ethnicity_classifer: https://github.com/BiagioAntonelli/Ethnicity_classifier
3. (Additional unexplored) Real-time face detection and classification: https://github.com/oarriaga/face_classification Please follow the detailed instructions in each repo to first recreate their results (download dataset: data.h5; download pretrained model, etc.)

Experimentation_emotion folder

1. notebooks:
   • Create biased datasets: as name of the notebook suggested, it creates biased datasets (exclude black in training, label black's emotion as neutral, label random people's emotion as neutral) based on data_race.h5.
   • Explore_fer2013: load race_predictions result and create data_race.h5 (including: image, emotion label and race prediction label) from data.h5

Experimentation_ethnicity folder

1. code_modified:

   • predict_array.py is a modified version of original predict.py (to save ethnicity predictions in pickle file)

2. data: -test_data: comes from data.h5

3. result

   • save pickled prediction results