Blender Data Generation

The main objective of this project is to provide an abstraction layer over Blender API in order to have a set of tools for an easy generation of labeled synthetic data.

Using it

In this repository an example for a basic generation is provided in "tinker.py". The blender project "YOLO_data_generator.blend" includes in the scripting section the script already opened.

You have to provide absolute path for:

• "dataset_path": where the dataset is located
• "render_img_saving_path": where we save our results

With EPOCHS we specify the number of images we want to generate

Roadmap

TODO:

• Render labeled image masks
• Option to make images look at the camera
• Option/params to scale images/+- scaling randomly (Class already done, needs to integrate)
• Make images "stand" on the floor as a parameter (Done by default)
• Add background simple
• Add 3D objects and handle them with DataHandler
• Scene rendering class
• Enable GPU by code
• Refactor code

FIX:

• Sometimes image planes are not deleted properly (probably related with class DataHandler with delete_images_planes())

Acknowledgments

• Federico Arenas López: For providing a good base project and functions for defining bounding boxes and rendering.