DStruct2Design: Data and Benchmarks for Data Structure Driven Generative Floor Plan Design

Paper

Our paper is available here

if you use this repository, please cite our work:

@misc{luo2024dstruct2designdatabenchmarksdata,
      title={DStruct2Design: Data and Benchmarks for Data Structure Driven Generative Floor Plan Design}, 
      author={Zhi Hao Luo and Luis Lara and Ge Ya Luo and Florian Golemo and Christopher Beckham and Christopher Pal},
      year={2024},
      eprint={2407.15723},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2407.15723}, 
}

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Prerequisites

In our paper, we train a LLama3-8B-Instruct model. Training is enabled by llama-receipe. You can either install llama-receipe, or install from requirement.txt

Install with llama-receipe:

pip install llama-recipes

Install from requirement.txt:

pip install -r requirements.txt

Datasets

ProcTHOR

You can download the converted ProcTHOR-10K dataset from here and put it under datasets/procthor_converted/

RPLAN

RPLAN dataset needs to be requested from its homepage.

After it's obtained, save all the data (pngs) under datasets/rplan/. Then run our conversion script to convert it. The converted dataset will be saved under datasets/rplan_converted/:

python scripts/rplan_dataset_convert.py

Pretrained Weights

The pretrained PEFT LoRA weights for all of our models can be obtained:

Weights for 4 model variants trained on RPLAN

https://drive.google.com/file/d/1cAYlEupNUGJefNdwkNaaq7fD3X3_P46D/view?usp=sharing

Weights for 3 bubble diagram model variants trained on ProcTHOR

https://drive.google.com/file/d/16cYPK6g_Ho4VbvjvBZIGHMzNTBWzcAZT/view?usp=drive_link

Weights for 3 constraint only (no bubble diagram) model variants trained on ProcTHOR

https://drive.google.com/file/d/13k-pBmhGhYthm4WbHzrRH7WjaSKNkTpq/view?usp=drive_link

After download, they can be un-compressed and put under their respective folder under models/.

Training

Alternatively, these weights can be trained from scratch with the following command:

to train on ProcTHOR:

python -m llama_recipes.finetuning --use_peft --peft_method lora --quantization --model_name meta-llama/Meta-Llama-3-8B-Instruct --batch_size_training 2 --num_epochs 8 --dataset "custom_dataset" --custom_dataset.file "procthor_dataset.py" --use_wandb False --wandb_config.project "floorplans" --output_dir procthor --exprm $EXPRM_VAR --load_peft False --ds_version $BD_VAR --load_peft False

here $BD_VAR and $EXPRM_VAR indicate the model variants to be trained as explained in Section 6.1 of our paper.

$BD_VAR can be set to either 'bd' or 'non_bd',

$EXPRM_VAR can be set to 'specific', 'mask', or 'preset_mask'

to train on RPLAN:

python -m llama_recipes.finetuning --use_peft --peft_method lora --quantization --model_name meta-llama/Meta-Llama-3-8B-Instruct --batch_size_training 2 --num_epochs 5 --dataset "custom_dataset" --custom_dataset.file "rplan_dataset.py" --use_wandb False --wandb_config.project "floorplans" --output_dir rplan --exprm $EXPRM_VAR$ --load_peft False

for RPLAN, the model variant is decided by just $EXPRM_VAR.

$EXPRM_VAR can be 1 of '5R', '6R', '7R', or '8R'. The differences between these variants are explained in Section 6.1 of our paper.

Inference

To run genneration after the pretrained weights are obtained, do the following:

(note that you can run greedy or sample generations. In our experiments, we use both, and sampling is done with num_samples set to 20).

To run generation on PROCTHOR-trained models:

python run_generation_procthor.py --exprm $EXPRM_VAR --num_samples 1 --version $BD_VAR

$BD_VAR can be set to either 'bd' or 'non_bd',

$EXPRM_VAR can be set to 'specific', 'mask', or 'preset_mask'

It will load the trained model variant according to the variable.

To run generation on RPLAN-trained models:

python run_generation_rplan.py --exprm $EXPRM_VAR --num_samples 1

$EXPRM_VAR can be 1 of '5R', '6R', '7R', or '8R'.

It will load the trained model variant according to the variable.

Evaluation

To evaluate generated results saved in $RESULTS_DIR, simply run the following command:

python run_metric.py $RESULTS_DIR