The verdict: Runway amazes me!
I’ve tried both, runway and ml5 and I think it’s safe to say that I’m a fan of runway and prefer it over ml5. I used im2txt, which is basically the same as ml5 image classifier on p5. I’ve showed both models a picture of my bun using the webcam, and the results were hilarious. I’ve noticed that runway is more accurate, however doesn’t show the confidence level which I like to see.
Ml5
My bun looked like a ‘window screen’ to the image classifier on p5 with a confidence level of 0.37. I had to take so many screenshots to actually read what it says because it’s too fast and I wanted to capture the item that got the highest level of confidence. I was sitting in a dim room and placed my webcam to only capture my bun. Therefore, the lighting wasn’t ideal to accurately capture anything that would get a high level of confidence. Also, I doubt people use pictures of my bun when teaching machines.
Runway
My bun looked like a ‘black cat sitting on top of a bed’ on runway. I can actually see it looking like a cat, but I’m not sure how confident runway is with that assumption. It also showed that my bun looked like ‘a woman in a black dress and a white shirt’ which is not accurate at all, but I liked that it detects color. In my opinion, detecting color can maybe trick the eye to see what the program is seeing because that way, the computer is assuring the user that it sees what’s in front of it.
Overall, I think runway has more potential and is more user friendly than ml5. That is because many people can use it for different purposes. For example, designers can edit pictures on runway to add filters without having anything to do with machine learning in mind. In addition, after looking at different sketches on ml5.js, some sketches feel repetitive to me somehow, as opposed to the variety of models you can find on runway.
But it’s safe to say that both programs are very powerful and can offer unlimited and powerful projects in the present and the future.
In my future projects, I can see myself using runway just because it doesn’t require coding and I’m currently still not confident with my coding skills. I don’t have a project idea so far; I still need to learn how far I can go and try many models. However, I see myself focusing on maybe embedding different cultures to it, for example, detect certain objects’ origin countries.
Some of my work on ml5:
The sketch above was done by Elena and I, using Kinect and Kinectron. It was our final project last semester for pcom.
Another sketch using ml5 with poseNet: https://editor.p5js.org/dinakhalil1/present/A9EtPAMb
Dedicated to my amazing and close friend Elena Glazkova
Some model I've tried on Runway
