Computer-implemented method for generating (training) images

An apparatus configured to generate labelled training images characterizing automatic robotic or manual manipulation of a plurality of stackable objects in a workspace, comprising:

• a communication interface;
• a processor, and
• a memory;
• wherein the communication interface is configured to obtain a first training image subset obtained at a first time index including a depth map and a visual image of a plurality of stackable objects in a stacking region of a workspace, and to obtain a second training image subset obtained at a second time index including a depth map and a visual image of the stacking region in the workspace, wherein the second training image subset characterizes a changed spatial state of the stacking region;
• and
• wherein the processor is configured to compute a depth difference mask based on the depth maps of the first and second training image subsets, to compute a visual difference mask based on the visual images of the first and second training image subsets, and to generate a segmentation mask using the depth difference mask and/or the visual difference mask, wherein the segmentation mask localizes a stackable object based on the spatial state of the stacking region at the first time index, before the spatial state was changed by automatic robotic or manual manipulation of at least one stackable object in the workspace.

A non-transitory machine-readable medium on which is stored a computer program including a set of machine readable instructions for generating labelled training images characterizing automatic robotic or manual manipulation of a plurality of stackable objects in a workspace, the instructions, when executed by a computer, causing the computer to perform the following steps:

• obtaining a first training image subset obtained at a first time index including a depth map and a visual image of a plurality of stackable objects in a stacking region of a workspace;
• obtaining a second training image subset obtained at a second time index including a depth map and a visual image of the stacking region in the workspace, wherein the second training image subset characterizes a changed spatial state of the stacking region;
• computing a depth difference mask based on the depth maps of the first and second training image subsets;
• computing a visual difference mask based on the visual images of the first and second training image subsets;
• generating an annotated segmentation mask using the depth difference mask and/or the visual difference mask, wherein the annotated segmentation mask localizes a stackable object based on a spatial state of the stacking region at the first time index, before the spatial state was changed by automatic robotic or manual manipulation of at least one stackable object in the workspace.