The Scene Language: Representing Scenes with Programs, Words, and Embeddings

arXiv | Project Page

Yunzhi Zhang, Zizhang Li, Matt Zhou, Shangzhe Wu, Jiajun Wu. arXiv preprint 2024.

Installation

Environment

conda create --name sclg python=3.11
conda activate sclg
pip install mitsuba 
# if you run into segmentation fault, you might need specific mitsuba versions
# e.g., `pip install --force-reinstall mitsuba==3.5.1` on MacOS
pip install unidecode Pillow anthropic transforms3d astor ipdb scipy jaxtyping imageio

# required for minecraft renderer
pip install spacy
python -m spacy download en_core_web_md

pip install --force-reinstall numpy==1.26.4  # to be compatible with transforms3d

git clone https://github.com/zzyunzhi/scene-language.git
cd scene-language
pip install -e .

Language Model API

Get your Anthropic API key following the official documentation and add it to engine/key.py:

ANTHROPIC_API_KEY = 'YOUR_ANTHROPIC_API_KEY'
OPENAI_API_KEY = 'YOUR_OPENAI_API_KEY'  # optional, required for `LLM_PROVIDER='gpt'`

By default, we use Claude 3.5 Sonnet. You may switch to other language models by setting LLM_PROVIDER in engine/constants.py.

Text-Conditioned 3D Generation

Renderer: Mitsuba

python scripts/run.py --tasks "a chessboard with a full set of chess pieces" 

Renderings will be saved to ${PROJ_ROOT}/scripts/outputs/run_${timestep}_${uuid}/${scene_name}_${uuid}/${sample_index}/renderings/*.gif.

Example results (raw outputs here):

"a chessboard with a full set of chess pieces"	"A 9x9 Sudoku board partially filled with numbers"	"a scene inspired by Egon Schiele"	"a Roman Colosseum"	"a spider puppet"

Renderer: Minecraft

ENGINE_MODE=minecraft python scripts/run.py --tasks "a detailed cylindrical medieval tower"

Generated scenes are saved as json files in ${PROJ_ROOT}/scripts/outputs/run_${timestep}_${uuid}/${scene_name}_${uuid}/${sample_index}/renderings/*.json. For visualization, run the following command:

python viewers/minecraft/run.py

Then open http://127.0.0.1:5001 in your browser and drag generated json files to the web page.

Example results (raw outputs here):

"a witch's house in Halloween"	"a detailed cylindrical medieval tower"	"a detailed model of Picachu"	"Stonehenge"	"a Greek temple"

Image-Conditioned 3D Generation

python scripts/run.py --tasks ./resources/examples/* --cond image --temperature 0.8

Codebase Details

Macro definitions

The following table lists helper functions defined in this file in accordance with expressions defined in the domain-specific language (DSL) (Tables 2 and 5 of the paper):

Implementation	DSL
register	bind
library_call	call
primitive_call	call
loop	union-loop
concat_shapes	union
transform_shape	transform
rotation_matrix	rotation
translation_matrix	translate
scale_matrix	scale
reflection_matrix	reflect
compute_shape_center	compute-shape-center
compute_shape_min	compute-shape-min
compute_shape_max	compute-shape-max
compute_shape_sizes	compute-shape-sizes

Codebase improvements

The current codebase allows you to generate 3D scenes with text or image prompts. Other tasks and renderers reported in the paper will be supported in future updates.

Please submit a PR or email us if you have feature requests, suggestions for improvements, or would like to share your results.

Citation

If you find this work useful, please consider cite our paper:

@article{zhang2024scenelanguage,
  title={The Scene Language: Representing Scenes with Programs, Words, and Embeddings},
  author={Yunzhi Zhang and Zizhang Li and Matt Zhou and Shangzhe Wu and Jiajun Wu},
  year={2024},
  journal={arXiv preprint arXiv:2410.16770},
}