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..
PlayerOne: Egocentric World Simulator
Authors: Yuanpeng Tu, Hao Luo, Xi Chen, Xiang Bai, Fan Wang, Hengshuang Zhao
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate its great generalization ability in precise control of varying human movements and world-consistent modeling of diverse scenarios. |
| Researcher Affiliation | Collaboration | 1HKU 2DAMO Academy, Alibaba Group 3Hupan Lab 4HUST |
| Pseudocode | No | The paper describes methodologies in detail with text and figures (e.g., Figure 2 and Figure 3), but it does not contain a dedicated section or block explicitly labeled as "Pseudocode" or "Algorithm". |
| Open Source Code | No | The paper includes a project website URL "https://playerone.github.io" in the header, which is typically a high-level overview page and not a direct link to a source code repository. While the NeurIPS checklist in the paper states 'Yes' for open access to data and code and refers to 'supplementary material', this document (which includes the supplementary material) does not provide a direct link to a code repository for the methodology described, nor an explicit statement of code release in the main text or appendix. |
| Open Datasets | Yes | Our training dataset combines multiple publicly available datasets to ensure comprehensive coverage of diverse environmental contexts, action types, and intensity levels, thereby enhancing model generalization. Table 1: Statistics of datasets used for training our Player One. [...] Appendix D License of assets: Datasets (Apache 2.0 License) Nymeria [21]/FT-HID [11]/Ego Exo-Fitness [18] (Creative Commons Attribution 4.0 International), Ego Exo4D [8]/Egovid-5M [33](MIT License). |
| Dataset Splits | Yes | Since there is no publicly available benchmark for our task, we construct a benchmark with 100 videos collected from Nymeria [21] dataset, which is not included for training. |
| Hardware Specification | Yes | We train our model for 100,000 steps on 8 NVIDIA A100 GPUs with a batch size of 56 and sample resolution of 480 480. |
| Software Dependencies | No | The paper mentions several models, techniques, and tools used (e.g., Wanx2.1 1.3B, Adam optimizer, Lo RA, SAM2, SMPLest-X, Open Pose, CUT3R, DUSt3R, Qwen2.5-VL), but it does not specify explicit version numbers for general software libraries or environments like Python, PyTorch, or CUDA, which are required for a reproducible software dependency description. |
| Experiment Setup | Yes | We set the Lo RA rank and the update weight of the matrices as 128 and 4 respectively and initialize its weight following [30]. The inference step and the learning rate are set as 50 and 1 × 10−5 respectively, where the Adam optimizer and mixed-precision bf16 are adopted. The cfg of 7.5 is used. We train our model for 100,000 steps on 8 NVIDIA A100 GPUs with a batch size of 56 and sample resolution of 480 × 480. The generated video runs at eight frames per second, and we utilize 49 video frames (6 seconds) for training. |