...Unlike voxel-based or point-based approaches, Mesh R-CNN uses a differentiable mesh representation, allowing it to efficiently refine surface geometry while maintaining high spatial detail. The system combines 2D detection from Mask R-CNN with 3D reasoning modules that output full mesh reconstructions aligned with the input image. It has been evaluated on datasets such as Pix3D, where it demonstrates state-of-the-art performance in reconstructing real-world object geometry.