Core Concepts
SKoPe3D introduces a synthetic dataset for vehicle keypoint perception in 3D, addressing challenges in traffic monitoring.
Abstract
SKoPe3D presents a synthetic dataset for vehicle keypoint perception in 3D, crucial for intelligent transportation systems. The dataset includes images with bounding boxes, tracking IDs, and 33 keypoints per vehicle. It spans over 25k images across 28 scenes, containing over 150k vehicle instances and 4.9 million keypoints. By leveraging this dataset, advancements in vehicle keypoint detection for ITS can be achieved. The data generation pipeline involves three stages: 3D annotation, scene configuration, and simulation. Evaluation metrics include PCK, precision-recall of bounding boxes and keypoints. Results show good performance on easy and medium scenes but challenges in hard scenes due to occlusion and weather conditions.
Stats
SKoPe3D contains over 150k vehicle instances and 4.9 million keypoints.
The training set consists of 10.5k images from ten scenes.
Testing data contains 3k images from three unseen scenes.
Quotes
"We propose SKoPe3D, a unique synthetic vehicle keypoint dataset generated using the CARLA simulator."
"Our experiments highlight the dataset’s applicability and the potential for knowledge transfer between synthetic and real-world data."