This paper introduces Group Normalization (GN) as an alternative to Batch Normalization (BN) in deep learning. While BN has been a milestone in deep learning, it suffers from issues when the batch size is small, leading to inaccurate statistics and increased model error. GN addresses this by normalizing features within groups of channels, making it independent of batch sizes and more stable across different batch sizes. The experiments show that GN outperforms BN, especially with small batch sizes, and is effective in various tasks such as object detection, segmentation, and video classification. GN is easy to implement and can be transferred from pre-training to fine-tuning. The paper also compares GN with other normalization methods like Layer Normalization (LN) and Instance Normalization (IN), showing that GN performs better in visual recognition tasks. The results demonstrate that GN is a strong alternative to BN, offering better performance and stability in a wide range of applications.This paper introduces Group Normalization (GN) as an alternative to Batch Normalization (BN) in deep learning. While BN has been a milestone in deep learning, it suffers from issues when the batch size is small, leading to inaccurate statistics and increased model error. GN addresses this by normalizing features within groups of channels, making it independent of batch sizes and more stable across different batch sizes. The experiments show that GN outperforms BN, especially with small batch sizes, and is effective in various tasks such as object detection, segmentation, and video classification. GN is easy to implement and can be transferred from pre-training to fine-tuning. The paper also compares GN with other normalization methods like Layer Normalization (LN) and Instance Normalization (IN), showing that GN performs better in visual recognition tasks. The results demonstrate that GN is a strong alternative to BN, offering better performance and stability in a wide range of applications.