Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
DeBaRA: Denoising-Based 3D Room Arrangement Generation
Authors: Léopold Maillard, Nicolas Sereyjol-Garros, Tom Durand, Maks Ovsjanikov
NeurIPS 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our approach through extensive experiments and demonstrate significant improvement upon state-of-the-art approaches in a range of scenarios. |
| Researcher Affiliation | Collaboration | Léopold Maillard1,2 Nicolas Sereyjol-Garros Tom Durand2 Maks Ovsjanikov1 1LIX, École Polytechnique, IP Paris 2Dassault Systèmes |
| Pseudocode | Yes | Algorithm 1 Self Score Evaluation |
| Open Source Code | No | Unfortunately, the code for De Ba RA cannot be disclosed due to author affiliation. |
| Open Datasets | Yes | Our experiments are conducted on the 3D-FRONT [10] synthetic indoor layouts, furnished with assets from 3D-FUTURE [11] that we use as the object retrieval database. |
| Dataset Splits | No | The paper states 'leading respectively to 2338/587 and 2071/516 train/test splits' but does not explicitly provide numerical details for a validation split for reproducibility. |
| Hardware Specification | Yes | All the training and evaluation experiments as well as the computation of generation times reported in Table 5 have been performed on a single NVIDIA RTX A6000 GPU. |
| Software Dependencies | No | While PyTorch and Meta Llama-3-8B are mentioned, specific version numbers for PyTorch and other key software components are not provided, preventing full reproducibility of software dependencies. |
| Experiment Setup | Yes | We trained our models separately on the 3D-FRONT [10] living room and dining room subsets for 3000 epochs, with a batch size of 32 and monitor the validation loss to avoid overfitting of the training set in the late iterations. We use the Adam W [29] optimizer with its Py Torch default parameters and learning rate η = 10 4, scheduled with a linear warmup phase for the first 50 epochs, starting at η 0.01. Following this, a cosine annealing schedule [28] reduces η to a minimum of 10 8 over 2200 epochs. |