AI Selfie Evaluator

Inspiration

Finding good, professional photos is a common problem, so we decided to tackled this issue.

What it does

You insert a photo URL, and our application rates the photo based on lighting, object spacial orientation, quality, and face obstruction. It returns an overall score which determines the overall quality of the photo.

How we built it

We used the Microsoft Custom Vision API to classify photos based on the criteria above. We then used Python and tkinter to create an interactive window where you can input your photo URL.

Challenges we ran into

The training portion in the custom computer vision interface presented a challenge, especially when we tried to improve precision and accuracy. We realized that we needed to select data points (images) that are diverse but still clearly enough "badly lit" or "good quality" or whatever else their tag required.

Accomplishments that we're proud of

We are proud of our trained computer vision model and how well it is able to detect image lighting, quality, positioning, and facial visibility.

What we learned

We learned a lot about machine learning/ computer vision strategies for training the computer, as well as the new technologies available that make it easier to implement and utilize these strategies as developers.

What's next for Selfie Evaluator

We hope to implement a feature such that the user can import an album or large number of url's to images, and have the program determine which is the "best" image.

Built With

• microsoft-custom-vision