Fetching the paper…

Image Representations Learned With Unsupervised Pre-Training Contain Human-like Biases · Around