Hierarchically-Structured Open-Vocabulary Indoor Scene Synthesis with Pre-trained Large Language Model
Authors: Weilin Sun, Xinran Li, Manyi Li, Kai Xu, Xiangxu Meng, Lei Meng
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive comparison experiments and ablation studies with both qualitative and quantitative evaluations to validate the effectiveness of our key designs with the hierarchically structured scene representation. Our approach can generate more reasonable scene layouts while better aligned with the user requirements and LLM descriptions. |
| Researcher Affiliation | Academia | 1School of Software, Shandong University, China 2 School of Computer Science, National University of Defense Technology, China 3 Shandong Research Institute of Industrial Technology, China EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes methods such as 'divide-and-conquer optimization' and formulations using mathematical notation, but it does not present any explicitly labeled pseudocode or algorithm blocks with structured steps. |
| Open Source Code | No | The paper mentions retraining deep networks using 'their released code' (referring to other methods) and using 'their implementations of the pipelines' for LLM-assisted approaches. However, there is no explicit statement or link indicating that the authors have released their own source code for the methodology described in this paper. |
| Open Datasets | Yes | We conduct the comparison and ablation study on the 3D-Front dataset (Fu et al. 2021). Finally, we retrieve 3D object models from Objaverse (Deitke et al. 2023) and 3D-Front datasets (Fu et al. 2021) based on the CLIP scores |
| Dataset Splits | Yes | The sizes of the training sets are 3397 and 690 for bedrooms and living rooms, while the corresponding test sets are 60 and 53. |
| Hardware Specification | Yes | The network is trained on the combination of the bedroom and living room training sets, which takes about 8 hours on a Nvidia 4090 GPU. |
| Software Dependencies | No | The paper mentions 'GPT-4(Achiam et al. 2023)' and 'GUROBI solver (Gurobi Optimization, LLC 2023)'. While these tools are identified, specific version numbers for the software themselves (e.g., GPT-4 version X, GUROBI version Y) are not provided, only citations to their reference papers or the company year of the manual. |
| Experiment Setup | Yes | We train the hierarchy-aware neural network with 500 epochs using the Adam optimizer, where the batch size is 4 and the learning rate is 1e-4. |