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..
Learning to Generate Human-Human-Object Interactions from Textual Descriptions
Authors: Jeonghyeon Na, Sangwon Baik, Inhee Lee, Junyoung Lee, Hanbyul Joo
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experiments 5.1 Baselines and Metrics 5.2 Results Tab. 1 shows quantitative results evaluating the realism on generating two people in action with object. Compared to baseline models, our model achieves significantly higher score in body pose and distance FD, implying our model can produce more realistic and natural HHOIs, that resemble those in real world environment. |
| Researcher Affiliation | Academia | Jeonghyeon Na , Sangwon Baik , Inhee Lee, Junyoung Lee, Hanbyul Joo Seoul National University Equal Contribution Corresponding Author EMAIL |
| Pseudocode | No | The paper describes the methodology using mathematical formulations and descriptive text, but it does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | We intend to release the code and dataset publicly after further polishing and preparation. However, they are not included at submission time, and we do not currently provide access to ensure the release meets quality and usability standards. |
| Open Datasets | Yes | The same data processing and training procedure is applied to the CORE4D dataset. ... We compare the distributions of generated results against the test set in CORE4D and our collected dataset. ... [37] Y. Liu, C. Zhang, R. Xing, B. Tang, B. Yang, and L. Yi. Core4d: A 4d human-object-human interaction dataset for collaborative object rearrangement. ar Xiv preprint ar Xiv:2406.19353, 2024. |
| Dataset Splits | No | The paper mentions dividing the collected HHOI dataset into HOI and HHI subsets for training, and evaluating on a 'test set in CORE4D and our collected dataset.' However, it does not provide specific percentages, sample counts, or detailed methodology for training/validation/test splits in the main text. |
| Hardware Specification | No | The main paper does not contain specific hardware details used for running experiments. The NeurIPS checklist indicates that these details are provided in the supplementary material. |
| Software Dependencies | Yes | We solve the ODE using external library [7], which is fully supported to run on the GPU. [7] R. T. Q. Chen. torchdiffeq, 2018. URL https://github.com/rtqichen/torchdiffeq. |
| Experiment Setup | Yes | We use SMPL-X [39] for the human model and 6D representation [70] for RH. ... We embed 126D human body poses in a 10D space, that is, H = 10, which results in ϕHOI R20 and ϕHHI R29. ... where λ1 and λ2 are weight terms. |