Mixed Diffusion Models for 3D Indoor Scene Synthesis

This repository contains the model code that accompanies our paper Mixed Diffusion for 3D Indoor Scene Synthesis. We present MiDiffusion, a novel mixed discrete-continuous diffusion model architecture, designed to synthesize plausible 3D indoor scenes from given room types, floor plans, and potentially pre-existing objects. Our approach uniquely implements structured corruption across the mixed discrete semantic and continuous geometric domains, resulting in a better conditioned problem for the reverse denoising step.

We place the preprocessing and evaluation scripts for the 3D-FRONT and 3D-FUTURE datasets based on ATISS in our ThreedFront dataset repository to facilitate comparisons with other 3D scene synthesis methods using the same datasets. ThreedFront also contains dataset class implementations as a standalone threed_front package, which is a dependency of this repository. We borrow code from VQ-Diffusion and DiffuScene for discrete and continuous domain diffusion implementations, respectively. Please refer to related licensing information in external_licenses.

If you found this work useful, please consider citing our paper:

@article{Hu24arxiv-MiDiffusion,
  author={Siyi Hu and Diego Martin Arroyo and Stephanie Debats and Fabian Manhardt and Luca Carlone and Federico Tombari},
  title={Mixed Diffusion for 3D Indoor Scene Synthesis},
  journal = {arXiv preprint: 2405.21066},
  pdf = {https://arxiv.org/abs/2405.21066},
  Year = {2024}
}

Installation & Dependencies

Our code is developed in Python 3.8 with PyTorch 1.12.1 and CUDA 11.3.

First, from this root directory, clone ThreedFront:

git clone git@github.com:MIT-SPARK/ThreedFront.git ../ThreedFront

You can either install all dependencies listed in ThreedFront, or, if you also want to use threed_front for other projects, install threed_front separately and add its site-packages directory. For example, if you use virtualenv, run

echo "<ThreedFront_venv_dir>/lib/python3.x/site-packages" > <MiDiffusion_venv_dir>/lib/python3.x/site-packages/threed-front.pth

Then install threed_front and midiffusion. midiffusion requires two additional dependencies: einops==0.8.0 and wandb==0.17.1.

# install threed-front
pip install -e ../ThreedFront

# install midiffusion
python setup.py build_ext --inplace
pip install -e .

Dataset

We use 3D-FRONT and 3D-FUTURE datasets for training and testing of our model. Please follow the data preprocessing steps in ThreedFront. We use the same data files as those included in ThreedFront/data_files for training and evaluation steps. Please check that PATH_TO_DATASET_FILES and PATH_TO_PROCESSED_DATA in scripts/utils.py are pointing to the right directories.

Training

To train diffuscene on 3D Front-bedrooms, you can run

python scripts/train_diffusion.py <path_to_config_file> --experiment_tag <experiment_name>

We provide example config files in the config/ directory. This train script saves a copy of the config file (as config.yaml) and log intermediate model weights to output/log/<experiment_name> unless --output_directory is set otherwise.

Experiment

The scripts/generate_results.py script can compute and pickle synthetic layouts generated by a trained model through the threed_front package.

python scripts/generate_results.py <path_to_model_file> --result_tag <result_name>

This script loads config from the config.yaml file in the same directory as <path_to_model_file> if not specified. The results will be saved to output/predicted_results/<result_name>/results.pkl unless --output_directory is set otherwise. We can run experiments with different object constraints using the same model by setting the --experiment argument. The options include:

• synthesis (default): scene synthesis problem given input floor plans.
• scene_completion: scene completion given floor plans and existing objects (specified via --n_known_objects).
• furniture_arrangement: scene completion given floor plans, object labels and sizes.
• object_conditioned: scene completion given floor plans, object labels.
• scene_completion_conditioned: scene completion given floor plans, existing objects, and labels of remaining objects.

You can then render the predicted layout to top-down projection images using scripts/render_results.py in ThreedFront for evaluation.

python ../ThreedFront/scripts/render_results.py output/predicted_results/<result_name>/results.pkl

Please read this script for rendering options.

Evaluation

The evaluation scripts in the scripts/ directory of ThreedFront include:

• evaluate_kl_divergence_object_category.py: Compute KL-divergence between ground-truth and synthesized object category distributions.
• compute_fid_scores.py: Compute average FID or KID (if run with "--compute_kid" flag) between ground-truth and synthesized layout images.
• synthetic_vs_real_classifier.py: Train image classifier to distinguish real and synthetic projection images, and compute average classification accuracy.
• bbox_analysis.py: Count the number of out-of-boundary object bounding boxes and compute pairwise bounding boxes IoU (this requires sampled floor plan boundary and normal points).

Video

An overview of MiDiffusion is available on YouTube:

Relevant Research

Please also check out the following papers that explore similar ideas:

• Fast and Flexible Indoor Scene Synthesis via Deep Convolutional Generative Models pdf
• Sceneformer: Indoor Scene Generation with Transformers pdf
• ATISS: Autoregressive Transformers for Indoor Scene Synthesis pdf
• Indoor Scene Generation from a Collection of Semantic-Segmented Depth Images pdf
• Scene Synthesis via Uncertainty-Driven Attribute Synchronization pdf
• LEGO-Net: Learning Regular Rearrangements of Objects in Rooms pdf
• DiffuScene: Denoising Diffusion Models for Generative Indoor Scene Synthesis pdf