As the name suggested, image cleanup is a feature within the listing flow that utilizes artificial intelligence to help sellers have a cleaner image. It is a very existing project for me as I’ve been wanting to work on something related to machine learning for a while.
The first phase of the project is an employee beta program. The goal was to test out and adjust the machine learning algorithms so it could become more precise before it was available to the public users. I worked closely with 2 engineers to help me learn more about the context and previous research done on similar topics. I also looked into similar services to learn about the flow and the control used for touchup.
In the end, we decided to opt for a auto-cleanup approach for the first phase of the project to reduce the scope and highlight the feature in the listing flow. During this phase, it was difficult to track the touchup record from users to feed it back to the algorithm. To help solicit some feedbacks, I proposed to utilize a survey that shows up after user declines the auto-cleanup photo, to get a gauge of the cleanup quality and the user experience.
After the phase 1 was made available to the employees, we continued to work on phase 2 of the project, which allows users to manually touch up the photos. Using designs from similar service as reference and the flow engineers had built out, I designed the flow into existing photo editing tools. To ensure the new flow is consistent with the design system, I also worked closely with our Patterns team on the logo and notification design. We also made some minor improvements to the tutorials based on feedbacks of phase 1 implementation.
The feature was officially launched in February 2020, and it has been well-received by our sellers so far.



