The way we consume video content has changed dramatically in recent years. Content originally produced for television in a horizontal format is now expected to be adapted to vertical formats such as 9:16 for mobile platforms like TikTok, Instagram, and YouTube Shorts.
Adapting video to these new formats is far from trivial. A simple center crop often removes the most important part of the scene, especially in sports and live events where the main subject moves rapidly and unpredictably.
In this talk, I will present AutoCrop, a deep learning–based system that automatically identifies regions of interest in videos to determine the optimal crop. Today, the system processes approximately 60% of the videos produced at WSC Sports, supporting more than 50 different sports, each with unique visual characteristics, motion patterns, and camera angles. I will share how we designed a system that works reliably across such a wide range of sports content.
Shiran Aziz is a Computer Vision and Deep Learning Algorithm Developer at WSC Sports, working on video understanding systems for detecting and tracking regions of interest in sports broadcasts. She holds a BSc in Mathematics and Computer Science and an MSc in Computer Science from the Hebrew University of Jerusalem. During her MSc, her thesis research focused on audio enhancement from multiple crowdsourced recordings, and this work was presented as an oral paper at Interspeech 2024. Her professional interests include computer vision, video understanding, audio processing, deep learning.
Keynote session: Hadas Grossmon Ella
Break
Lightning talks session
Roundtable closing
Talk by Hila Paz
Talk by Dr. Moran Mizrahi
Closing remarks
End
