This week, I use BodyPix and PoseNet to develop my interactive portraiture. BodyPix is used to recognized and draw body segmentation with random colors, which is pixelated with iteration. And PoseNet is used to recognize the position of right wrist that triggers the change of pixels’ colors.
Here is the image of pixelated body segments that I draw with BodyPix with an array that match the color of each segments.
And here is the video of final achievement.
I also tried to resize the canvas and the image so that they match the width and height of the window. However, it became too slow for people to interact with. And the models become too insensitive to detect the wrist’s position. I think it may result from the massive number of pixels of a full-screen image, which makes it too much information for the model to analyze at a time. So I changed back to with width of 320 and height of 240.