CommonScenes: Generating Commonsense 3D Indoor Scenes with Scene Graph Diffusion

Authors: Guangyao Zhai, Evin Pınar Örnek, Shun-Cheng Wu, Yan Di, Federico Tombari, Nassir Navab, Benjamin Busam

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments are conducted on SG-FRONT, where Common Scenes shows clear advantages over other methods regarding generation consistency, quality, and diversity.
Researcher Affiliation Collaboration 1Technical University of Munich 2Munich Center for Machine Learning 3Google
Pseudocode No The paper does not contain a section or figure explicitly labeled "Pseudocode" or "Algorithm," nor does it present structured code-like blocks outlining a procedure.
Open Source Code Yes Codes and the dataset are available on the website. https://sites.google.com/view/commonscenes
Open Datasets Yes Due to the lack of scene graph datasets also providing high-quality object meshes, we construct SG-FRONT, a set of well-annotated scene graph labels, based on a 3D synthetic dataset 3D-FRONT [18] that offers professionally decorated household scenarios. ... Codes and the dataset are available on the website.
Dataset Splits No The paper mentions that "SG-FRONT comprises around 45K 3D samples" and that "Extensive experiments are conducted on SG-FRONT," but it does not explicitly specify the training, validation, and test splits (e.g., percentages, sample counts, or predefined split references) needed for reproduction.
Hardware Specification Yes We conduct the training, evaluation, and visualization of Common Scenes on a single NVIDIA A100 GPU with 40GB memory.
Software Dependencies No The paper mentions using "Adam W optimizer" but does not specify other key software dependencies such as libraries (e.g., PyTorch, TensorFlow) or programming languages with their version numbers.
Experiment Setup Yes We set {λ1, λ2, λ3} = {1.0, 1.0, 1.0} in all our experiments. Nc in distribution Z is set to 128 and TSDF size D is set as 64.