Big Data Assisted Object Detection with Privacy Protection

The issues of privacy and bias in datasets are rapidly becoming important challenges that the computer vision field needs to address. So far, there has been little attention paid to solutions for protecting the privacy of new datasets. In our work, we explored a object detection solution on the WIDER FACE dataset by anonymizing the dataset using face synthesis and enhancing the WIDER FACE dataset by balancing facial features along the dimensions of gender and skin color. Using both the original dataset and our enhanced dataset to train the target detection model, our target detection results show that our model can maintain detection performance while preserving privacy and partially balancing bias.