The paper introduces SDDGR (Stable Diffusion-based Deep Generative Replay), a novel approach for class incremental object detection (CIOD) that leverages stable diffusion models to generate realistic and diverse synthetic images. SDDGR addresses the challenge of catastrophic forgetting in CIOD by incorporating an iterative refinement strategy and L2 knowledge distillation to improve the retention of prior knowledge. The method also includes pseudo-labeling to prevent misclassification of old objects in new task images. Extensive experiments on the COCO 2017 dataset demonstrate that SDDGR significantly outperforms existing algorithms, achieving state-of-the-art performance in various CIOD scenarios. The key contributions of the paper include the application of diffusion-based generative models in CIOD, the improvement of image quality through iterative refinement, and the effective use of synthetic images for training.The paper introduces SDDGR (Stable Diffusion-based Deep Generative Replay), a novel approach for class incremental object detection (CIOD) that leverages stable diffusion models to generate realistic and diverse synthetic images. SDDGR addresses the challenge of catastrophic forgetting in CIOD by incorporating an iterative refinement strategy and L2 knowledge distillation to improve the retention of prior knowledge. The method also includes pseudo-labeling to prevent misclassification of old objects in new task images. Extensive experiments on the COCO 2017 dataset demonstrate that SDDGR significantly outperforms existing algorithms, achieving state-of-the-art performance in various CIOD scenarios. The key contributions of the paper include the application of diffusion-based generative models in CIOD, the improvement of image quality through iterative refinement, and the effective use of synthetic images for training.