Two Minute Papers explores the paper “ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness ” The source documents and code are available at the following links:
Two Minute Papers has a look at the paper “3D Human Pose Machines with Self-supervised Learning” in his latest video. This topic actually came up in a meeting on Friday.
In this video, Two Minute Papers examines the paper “DeepFocus: Learned Image Synthesis for Computational Displays” from Facebook Research. Having a neural network synthesize depth and focus information from a two dimensional input can make more content usable for virtual reality or mixed reality uses.
Two Minute Papers has posted a video exploring the paper “A Variational U-Net for Conditional Appearance and Shape Generation”