ART-InvRec: An adversarial framework for rotation-invariant 3D object reconstruction

Xinyan Yang, Rui Yang, Fei Hu, Li Fang, Wei Zhong, Zhiye Chen, George Wang

Abstract: Highlights•A novel adversarial framework ART-InvRec is proposed. It firstly demonstrates the feasibility of achieving rotation-invariant 3D object reconstruction through adversarial training.•ART-InvRec treats point cloud rotation as an adversarial attack and defends it by retraining the network on rotated inputs.•The research addresses the limitation of current 3D object reconstruction methods that primarily focus on the precision of aligned objects. This is particularly important in real-world applications like indoor navigation, self-driving cars, and robotics, where sensor-collected point cloud data may not always be aligned.