git clone https://github.com/milmillin/d4descent.git
cd d4descent
uv sync

Running the experiments requires two steps: generating the shell scripts for the experiments and running them. This can be done with the following commands:

cd runs
uv run python _rungen_arclines.py
./_generated/arclines/AL-F_OneComp.sh

The _rungen_*.py will generate the shell scripts for the experiments. It will print all the generated shell scripts which you can then run. Three grammars are tested in the paper: arclines for Arclines, trees for Tree, and ur for UnionRect grammars. The suffixes denote objective functions: _topopt for topology optimization, _prompts for optimizing with SDS, _pngs for optimzing towards images. No suffix means optimizing towards shapes.

For ablations, uncomment parts of the code in _rungen_*.py and rerun to regenerate the shell scripts.

The repository is organized as follows:

• src/d4descent/ - Main package containing the optimization framework
  • optimizer.py - Core optimization algorithms and procedures
  • scheduler.py - Learning rate scheduling and optimization control
  • context.py - Execution context management
  • metrics.py - Evaluation metrics and logging
  • visualizer.py - Visualization tools for results and debugging
  • object_collection.py - Management of optimizable objects
  • types.py - Type definitions and interfaces
  • util.py - Utility functions and helpers

• src/d4descent/objects/ - Implementation of shape grammar objects
  • arclines.py - Arclines grammar implementation with curves and connectors
  • tree.py - Tree grammar for hierarchical branching structures
  • ur.py - UnionRect grammar for axis-aligned rectangle compositions
• src/d4descent/tasks/ - Task-specific implementations for different optimization objectives
  • _base.py - Base task interface and common functionality
  • arclines.py - Arclines-specific optimization tasks
  • tree.py - Tree grammar optimization tasks
  • ur.py - UnionRect grammar optimization tasks
• src/d4descent/losses/ - Loss functions for various optimization targets
  • _base.py - Base loss function interface
  • raster.py - Rasterization-based losses for image targets
  • sds.py - Score Distillation Sampling loss for text-to-shape
  • topopt.py - Topology optimization losses for structural design
• src/d4descent/third_party/ - Third-party implementations and utilities
  • pid.py - PID controller implementation
  • sds.py - SDS-specific utilities and helpers
  • topopt.py - Topology optimization utilities

• configs/ - YAML configuration files for different experiments
  • losses/ - Loss function configurations
  • tasks/ - Task-specific configurations
• runs/ - Experiment generation and execution scripts
  • _rungen_*.py - Scripts to generate shell scripts for different grammars
  • _generated/ - Auto-generated shell scripts for experiments

• scripts/ - Standalone optimization scripts for specific tasks
  • optimize_pngs.py - Image-based optimization
  • optimize_prompts.py - SDS prompt optimization
  • optimize_shc.py - Shape optimization
• data/ - Input data for experiments
  • arclines/ - Arclines target shapes
  • pngs/ - Target images for optimization

• output/ - Results from optimization runs (generated during execution)

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

You are free to:

• Share — copy and redistribute the material in any medium or format
• Adapt — remix, transform, and build upon the material

Under the following terms:

• Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made.
• NonCommercial — You may not use the material for commercial purposes.