uvHolographics

uvHolographics is a small Blender add-on that helps you generate defect segmentation masks to enrich deep learning datasets. This might be useful in cases where there is a small amount of data, a heavy class imbalance, or the need to introduce specific instances that do not yet exist in the real world.

Basic example: ball with scratches.

Advanced example: motorcycle helmet with dents and scratches.

Requirements

• Blender >=2.82

Quickstart

1. Create a new file.
2. Run the add-on script.
3. Set up the basic material.
4. Apply the defects.
5. Finetune the material.
6. Check results.
7. Generate instances.

Instructional Videos

1. Basic usage: https://www.youtube.com/watch?v=eZwOeBkLL8E
2. Advanced case: motorcycle helmet: https://www.youtube.com/watch?v=iOIwj4IIdZU

Deep Learning Demo

Coming.

Future ideas

• Actually make it into a real add-on.
• Add functionality for multiple defect layers.
• Add a high level modeSwitcher node.

Relevant literature and other projects

• BlenderProc: A procedural blender pipeline to generate images for deep learning
  • github
• [Tabernik2019] Segmentation-Based Deep-Learning Approach for Surface-Defect Detection
  • arxiv
• [Shorten2019] A survey on Image Data Augmentation for Deep Learning
  • springeropen
• [Shrivastava2017] Learning from Simulated and Unsupervised Images through Adversarial Training
  • arxiv