Claire
What questions do you still have about the model and the associated data? Are there elements you would propose including in the biography?
This article talks about the bias that probably exist in the dataset for training the ml5 model. I wonder that when something has been updated or we want to correct the bias in the dataset, can we completely erase the previous memory of the model and input new cognition to it? Because once us, as a human, formed a stereotype of something, it’s a long and difficult process to change it. Is it also the case for machine learning model, or it’s easier to “convince” a machine?
How does understanding the provenance of the model and its data inform your creative process?
I may pay more attention on the sources and the possible bias that may contains in the model. In the creative process, I may avoid the controversial content.
https://editor.p5js.org/Claire_Z/sketches/cH9BnLCcp
I call this project “Take your umbrella”. It’s a rainy day, raise up your hands to take out the umbrella.
I use both BodyPose and BodySegmentation in this p5 sketch. I use BodySegmentation to mask the background and replace it with a rainy video. Then I use BodyPose to detect the left or right wrist. If the confidence of either wrist is greater than 0.3, the umbrella will appear. And I prepared two piece of audio. One is raining sound, and the other one is the sound of rain on umbrella. It begins with the plain rain sound, when you hold the umbrella, the audio will also changed to the rain on umbrella.
Since there is no points for hands in BodyPost, I use the position of wrists as the position for the umbrella, and shift it to make the umbrella looks like be holding on your hand.
But I still have an unsolved problem. In the first place, I wanted to use a picture of umbrella with transparent background. But when I preload the image, the program says failed to fetch. The image is available and the path I think is also correct. So finally, I use text and put an emoji umbrella there.