Overcome Racial Bias With Unity Vision
Our app directly addresses facial recognition biases by empowering African Americans to contribute to the training of equitable, unbiased AI models.
Our app directly addresses facial recognition biases by empowering African Americans to contribute to the training of equitable, unbiased AI models.
Facial recognition technology exhibits bias, particularly in accurately identifying African American individuals, leading to higher rates of false positives and negatives. This issue, rooted in non-diverse training datasets, raises concerns about fairness and discrimination, undermining trust in its universality and effectiveness.
20% of the part
40% of the part
20% of the part
20% of the part
20% of the part
20% of the part
Gather a diverse dataset of facial images.
Clean, resize, normalize, and augment the dataset.
Extract meaningful facial features.
Choose a facial recognition model.
Train the model on labeled data.
Fine-tune model hyperparameters.
Detect faces within images or video frames.
Allows us to set specific criteria or metrics to determine whether a recognition match is accepted or rejected.
Our the model is initially trained on a broad dataset, and then, our dataset is introduced during the fine-tuning process. This step enables the model to learn and adapt to the unique features and characteristics of African American faces.
Ipsum dolor sit, amet consectetur adipisicing elit. Blanditiis at dicta debitis illum, veritatis repellendus maiores reiciendis consequatur suscipit dignissimos!
Ipsum dolor sit, amet consectetur adipisicing elit. Blanditiis at dicta debitis illum, veritatis repellendus maiores reiciendis consequatur suscipit dignissimos!
Ipsum dolor sit, amet consectetur adipisicing elit. Blanditiis at dicta debitis illum, veritatis repellendus maiores reiciendis consequatur suscipit dignissimos!
Ipsum dolor sit, amet consectetur adipisicing elit. Blanditiis at dicta debitis illum, veritatis repellendus maiores reiciendis consequatur suscipit dignissimos!
Ipsum dolor sit, amet consectetur adipisicing elit. Blanditiis at dicta debitis illum, veritatis repellendus maiores reiciendis consequatur suscipit dignissimos!
Ipsum dolor sit, amet consectetur adipisicing elit. Blanditiis at dicta debitis illum, veritatis repellendus maiores reiciendis consequatur suscipit dignissimos!
We listened to more than 3 hours of US Congress testimony on facial recognition so you didn't have to go through it.
MIT grad student Joy Buolamwini was working with facial analysis software when she noticed a problem: the software didn't detect her face -- because the people who coded the algorithm hadn't taught it to identify a broad range of skin tones and facial structures.
