Instance Augmentation

Instance augmentation is a data augmentation technique focused on enhancing datasets by modifying or generating entire objects or instances within images or other data modalities, rather than just individual pixels. Current research emphasizes developing automated methods, often leveraging generative models like diffusion models or employing adversarial training, to create diverse and realistic augmented instances without manual annotation. This approach is proving valuable across various applications, including improving the performance of deep learning models for tasks such as object detection, segmentation, and few-shot learning, as well as enhancing the robustness of models to adversarial attacks and addressing class imbalances in datasets.