DiDiGen - Blender Addon

The Blender addon named "DiDiGen" comes with a Blender file and the addon to create images of randomized dining rooms and corresponding binary masks of visible clean and soiled plates in the images with the aim to create synthetic randomized datasets for training image segmentation models. Furthermore, a .csv-file will be created with additional information about each rendered scene such as the used materials, amount of forks on the table, etc.

Requirement

For Usage

• Blender 4.2.1
• Blender Addon: Photographer 5.4.5
  • https://chafouin.gumroad.com/l/photographer5
• The Blender file: "dining_scene_render.blend"
• The Addon .zip file: "didigen_addon.zip"

For Development

• Blender 4.2.1

• Visual Studio Code

• VS Extension: "Blender Development" by Jacques Lucke

• Developed with Python 3.10.11 64-bit

• Helpful Links

  • Windows: https://www.youtube.com/watch?v=YUytEtaVrrc
  • Mac: https://www.youtube.com/watch?v=_0srGXAzBZE
  • Linux: https://www.youtube.com/watch?v=zP0s1i9EXeM

Installation

• Blender 4.2.1: https://www.blender.org/download/lts/4-2/

• For the Photographer and Dining Room Generator Addon

  • Open Blender.
  • On the top left, open the tab Edit > Preferences...
  • In the new opened Window called Blender Preferences, open the Add-ons tab on the left.
  • On the top right, click on the downward pointing arrow and select Install from Disk...
  • Navigate to the git/addon folder and select didigen.zip.
  • The addon Dining Room Generator is activated if the checkmark is activated, next to the addon name, in the list of addons under the Add-ons tab.

Settings

• Changing the Render Settings, so the CPU and GPU is used for rendering
  • Open Blender.
  • On the top left, open the tab Edit > Preferences...
  • In the new opened Window called Blender Preferences, open the System tab on the left.
  • Under the first tab named Cycles Render Devices, select OptiX or CUDA.
  • Enable the listed CPU and GPU by clicking on the checkboxes left from the GPU and CPU names.

Usage

After activating the addon Dining Room Generator, it can be found in the Viewport, which is the window that shows the 3D scene, on the bar on the right border. It might be hidden and can be opened by pressing N or clicking on the small leftward pointing arrow, which opens the panel of the addon.

How to render images

• Requirements:
  • The addons Photographer and Dining Room Generator must be activated.
  • The blender scene dining_scene_render.blend must be opened in Blender.
  • The object room in the collection Render Collection must be activated and visible in the 3D Viewport Window.

• Select the export folder for the rendered images, the corresponding binary masks and the datalogging csv file by pressing the Set Export Path button
  • For confirmation, the path of the export folder will be listed in the addon window.
• Set the starting index of the render images with the Start Image Index option
  • This option is important for the naming convention of the rendered images and the index in the .csv-file.
• The Amount of Images option sets how many images should be rendered in this session.
• Optional: Change the name of the csv-file that documents information of each rendered scene
• At last, press the Render images button and confirm the start of the rendering option.
  • Every customization option of the dining room scene will be automatically randomized. Afterwards, the rendering loop will be paused for twenty seconds and a preview of the scene will be rendered to determine the correct camera exposure with the Photographer-addon. After twenty seconds, the real rendering proces will start.

• Warning
  • During Rendering, please do not interact with the opened Blender scene to guarantee that the plugin is working.
  • To cancel the current rendering session, Blender must be closed and restarted, because the addon is in a rendering loop until all images are rendered.

Dining Room Objects

The scene contains different collections, which can be seen on the top right in Blender. The collections contains the different objects that make up the dining room scene.

• Room

  • The Room-collection consist of the original chair, floor, table and wall object. Selecting the chair or table object will open the Chair Characteristics or Table Characteristics-tab in the Dining Room Generator-Panel. Opening the tabs will show the different customization options for the table and chairs of the dining room.

• Render Collection

  • This collection containts a single room object, which is used for rendering the dining room scene. Selecting the room object opens the Room Characteristics-tab in the Dining Room Generator-panel. The tab allows to customize the dining room characteristics.

• Scene

  • The Scene-collection contains the distributed light object for indoor lighting, the camera object and the focus point for the camera. The camera object can be selected to customize camera settings.

  

• Napkins

  • The Napkins-collection contains the different napkins objects that are distributed on the plates and the table. Napkins are either ball-shaped, slightly crumpled or folded. For each state, there exist five objects that are randomly crumpbled or folded.

• Plate Distribution Link Objects

  • The Plate Distribution Link Objects-collection contains links/references to objects that are placed on top of the plates.

• Table Distribution Objects

  • The Table Distribution Objects-collection contains objects that are randomly placed on top of the dining room table.
    • Distractors-collection contains five objects that are randomly shaped and have random generated materials that are used as distractor objects.
    • Plate Alternatives-collection contains the same plate with different objects placed on top of it for more complexity in the dining room scene.
    • Tableware-collection contains a distractor-, fork-, glass-, knife-, napkin-, plate-, spoon-object. Selecting the fork-, glass-, knife-, plate- or spoon-object opens the characteristics tab for each object in the Dining Room Generator-panel. The tab allows to customize the characteristics of each object.
    • Napkin Objects-collection contains links/references to objects that are placed on top of the table.

In case an option or characteristic is not visible, the Modifier Properties-tab can be opened for more details.

Additionally, the implemented technology can be inspected by changing the workspace in Blender from Layout-workspace to Geometry Nodes-workspace on the top of the Blender window.

Materials

Materials of each object can be further customized (such as glossiness or metalness) going to the Material Properties-panel.

The implemented technology can be inspected by changing the workspace in Blender from Layout-workspace to Shading-workspace on the top of the Blender window. Each material was developed with Shader Nodes.

Further Notes

• Compositing-Nodes under the Composting-workspace was used implement the rendering of the image and the corresponding binary mask.
• The implementation of the environment lighting can be found in the Shading-workspace and changing the Shader Type from object to world.

Acknowledgments

This work was developed for the scientific publication Philipp Mach, Tim Laue, Arne Hasselbring and Udo Frese: DiDiGen - Dirty Dishes Generator for Randomized Visual Training Data. In: RoboCup 2025: Robot World Cup XXVIII. Springer, to appear.

The research reported in this paper has been partially supported by the German Research Foundation DFG, as a part of Collaborative Research Center (Sonderforschungsbereich) 1320 Project-ID 329551904 EASE - Everyday Activity Science and Engineering, University of Bremen. The research was conducted in the subproject R02 Multi-cue perception supported by background knowledge.

Images generated by DiDiGen as well as neural networks trained by using these images are provided online.

DiDiGen was originally developed for the master thesis "I Cannot Believe All this Dirt! - Analysis of Synthetic Soil-Based Occlusion on Tableware Detectors with a Procedural Tableware Data Generator" by Philipp Mach in the computer science program of the University of Bremen.

Most, if not all, procedural-generated materials are heavily inspired by the YouTube channel Ryan King Art.