Image matching is a fundamental task in computer vision, and in recent years, deep learning-based matching models have gradually gained popularity. To address the issue of generalization in deep learning methods, researchers from Xiamen University, Intel, and DJI have proposed GIM: Learning Generalizable Image Matcher from Internet Videos. GIM enables matching models to acquire strong generalization capabilities from internet videos, making it suitable for training all matching models. The authors introduced the first Zero-shot Evaluation Benchmark (ZEB), and evaluation results show that GIM significantly enhances the generalization performance of matching models.