i2c-net: Using Instance-Level Neural Networks for Monocular Category-Level 6D Pose Estimation

Object detection and pose estimation are strict requirements for many robotic grasping manipulation applications to endow robots with the ability grasp objects different properties in cluttered scenes various lighting conditions. This work proposes framework i2c-net extract 6D of multiple belonging categories, starting from an instance-level network relying only on RGB images. The is trained a custom-made synthetic photo-realistic dataset, generated some base CAD models, opportunely deformed, enriched real textures domain randomization purposes. At inference time, employed combination 3D mesh reconstruction module, achieving category-level capabilities. Depth information used post-processing as correction. Tests conducted YCB-V NOCS-REAL datasets outline high accuracy proposed approach.