Скачать или смотреть Recall Isn’t All: Deep Dive into Model Evaluation Metrics - Liad Pollak Zuckerman

Join us for an engaging discussion on a topic often overlooked: model evaluations. Evaluation metrics play a crucial role in communicating with other groups, model comparison, setting product requirements and more. Using a parking slot detector as our primary example, we’ll talk about the difference between metrics used in papers and metrics for a model in production. We’ll dive into how we evaluate our models, even when specific parking slot type is filtered out at the product level. Additionally, we’ll address the process of selecting the best model from a set of ablations, provide guidance on conducting a fair comparison between two models. Furthermore, we will present our methodology for a unified scale to combine different metrics and measure our overall performance.

Liad Pollak Zuckerman - Senior Computer Vision Engineer
Liad is a senior computer vision researcher with over a decade of experience in deep learning and algorithm development for real-world applications. She holds an M.Sc. in Computer Science from the Weizmann Institute of Science and a B.Sc. in Computer Engineering from the Technion – Israel Institute of Technology. Her research has been published in top venues including ECCV and ICASSP.

hayaData 2025, Tel-Aviv, Israel
https://www.haya-data.com/