Training Image Generation Using Difference Masks for Object Segmentation

The system performs self-service by automatically generating training data through robotic manipulation actions. The robotic manipulator autonomously performs pick-and-place operations on objects, while the imaging system captures depth maps and visual images before and after manipulation. The processing system automatically computes difference masks and generates annotated segmentation masks without human intervention, enabling the system to create its own training data independently.

The system creates copies of physical manipulation scenarios in the form of digital training data. By capturing depth maps and visual images during robotic manipulation and processing them into annotated segmentation masks, the system generates synthetic training examples that replicate real-world object manipulation scenarios, enabling the machine learning model to learn from these copied experiences.

The system replaces the mechanical process of manual annotation by human operators with an automated computational process. Instead of human operators visually inspecting and annotating images, the system uses image processing algorithms to compute depth difference masks and visual difference masks, which are then processed into annotated segmentation masks. This substitution of mechanical human labor with automated computational methods maintains segmentation accuracy while achieving full automation.

The system introduces intermediary computational elements (depth difference masks and visual difference masks) that mediate between the raw image data and the final segmentation annotations. These intermediate representations serve as bridges that automatically translate physical manipulation scenarios into structured training data, eliminating the need for direct human annotation while preserving segmentation quality.

The system transitions from handling objects on a two-dimensional flat surface to handling objects in three-dimensional stacked configurations. By using depth cameras to capture depth maps and computing depth difference masks, the system gains the ability to perceive and manipulate objects in the vertical dimension, enabling it to handle densely stacked objects while maintaining ease of operation through automated robotic manipulation.

The system changes the spatial parameters of object arrangement from scattered two-dimensional placement to densely packed three-dimensional stacking. The robotic manipulator is designed to handle objects in various configurations, and the imaging system captures depth information that enables the system to adapt to different stacking densities and arrangements, thereby increasing versatility while maintaining operational ease through automation.