Meta develops AI that animates human-like figures
Meta, formerly Facebook, has developed an AI system that can identify and animate human-like figures in children’s drawings without any human guidance. This is a first-of-its-kind method that can automatically animate figures that come in many forms, colors, sizes, and scales with little morphology, body symmetry, and point of view.
To use this feature, parents have to upload the artwork to the system, and they can experience the excitement of watching their children’s drawing taking shape and becoming a moving character that can skip, jump, and dance. These drawings are downloadable and can also be submitted to improve the AI model.
The first step in animating the drawings is distinguishing the human figures from the background and other types of characters. Since the object detection technology lacks accurate animation segmentation, Meta researchers used bounding boxes, image processing, and morphological operations to obtain masks.
Read more: OpenAI Improves the Factual Accuracy Of GPT-3 Language Models
“We’re excited to announce a first-of-its-kind method for automatically animating children’s hand-drawn figures of people and humanlike characters (a character with two arms, two legs, a head, etc.) that bring these drawings to life in a matter of minutes using AI,” Meta researchers said in a statement.
They also used Meta’s convolutional neural network-based object detection model, Mask R-CNN, for extracting human-like characters from drawings. This model is pre-trained on the most extensive publicly available segmentation data set, made of photos of real-world objects rather than drawings.
Meta invited their colleagues to share and animate their kids’ artwork to train the AI with inanimate objects, obtaining approximately 1000 drawings. The researchers look forward to an AI system that can instantly create a detailed animated cartoon from complex drawings.