Class Notes

• you can use either CPU / GPU
  • ml5.js uses web browser GPU (WEBGL) as a default
  • but you can set backend(”CPU”) → if you have little input/little output model because it reduces the overhead caused to direct data to/from GPU
• map function
• COCO: was hand annotated!

Try

• segment anything으로 누끼따기!
• object를 segment한 다음에 해당 태그의 다른 이미지로 바꾸거나 generate해도 되나?

Learning Materials

• Computer Vision for Artists and Designers. ⇒ It sees what we can’t see and sees less than we can see.
• https://www.v7labs.com/blog/image-segmentation-guide
• Experiment with the provided code examples using ml5.js, TensorFlow.js, or Transformers.js. Try applying the models to different images and videos. Can you think of a creative way to integrate image segmentation into an interactive project?
• Document your experiments and share it on the Assignment 3 Wiki.

Work Desk

• I used to work with ML engineers in developing product features in an E-commerce app as product manager. We used segmentation models in defining product categories (trained on our data) and providing product suggestions.
• The word ‘prediction’
• What machines can see and humans can’t see - as we are different species, after all.

1. How many people are in this elevator?

   1. depth estimation → how many depths there are

   2. guess population within this area → calculate how many people there are in this area