yolibernal/causal_images

Causal Images

Using BlenderProc to create training images from a Structural Causal Model (SCM).

Setup

The package must be installed into Blender's addons, for example with

pip install . -t ./Blender/3.3/scripts/addons/

Example

Run

Sample from the SCM and render:

blenderproc run main.py sample --output_dir outputs/basic --scene_num_samples=5 --sampling_config examples/basic/sampling.json --fixed_config examples/basic/config.json

Generate counterfactuals from previous output:

blenderproc run main.py counterfactual --input_dir outputs/basic --output_dir outputs/basic_counterfactual --interventions_path examples/counterfactual/interventions.py

Export output as data/labels:

blenderproc run main.py export --input_dir outputs/basic --output_dir outputs/basic_export --to_image

Visualize

Visualize a generated image:

blenderproc vis hdf5 outputs/basic/0/0.hdf5