Shir Bar - Lecture 10 - Experimental Design in Computer Vision

This lecture explains how we can ensure that comparing methods is fair. PhD Candidate Shir Bar will also show how we can prove progress by building controlled experiments and ablation studies.
⛆ Contents ⛆
💻 Introduction
💻 Experiment design and hypothesis testing in computer vision
💻 Fair and unfair comparisons
💻 Controlled experiments and ablation studies
💻 Reporting uncertainty
💻 Evaluating model improvements in loss function, augmentations, training data, etc.
💻 Managing randomness in training
💻 Questions