Enhancing Synthetic Data Realism for Autonomous Vehicles Using Segmentation-Guided ControlNet

This dataset supports the paper "Enhancing Synthetic Data Realism for Autonomous Vehicles Using Segmentation-Guided ControlNet." It includes a refined version of the Virtual KITTI (VKITTI) dataset generated using a Stable Diffusion model fine-tuned on KITTI scenes with LoRA and guided by ControlNet using ground-truth segmentation maps and Canny edges. The dataset contains: Refined RGB images Aligned ground-truth segmentation maps Aligned depth maps This dataset was used to evaluate downstream tasks including object detection and depth estimation, showing significant improvements over the original synthetic data. GitHub repository: https://github.com/IqraNosheen786/Enhancing-Synthetic-Data-Realism-Using-Segmentation-Guided-ControlNet