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..
Deep Compositional Phase Diffusion for Long Motion Sequence Generation
Authors: Ho Yin Au, Jie Chen, Junkun Jiang, Jingyu Xiang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate the competitive performance of our proposed framework in generating compositional motion sequences that align semantically with the input conditions, while preserving phase transitional continuity between preceding and succeeding motion clips. Additionally, motion inbetweening task is made possible by keeping the phase parameter of the input motion sequences fixed throughout the diffusion process, showcasing the potential for extending the proposed framework to accommodate various application scenarios. Codes are available at https://github.com/asdryau/Trans Phase. |
| Researcher Affiliation | Academia | Ho Yin Au Hong Kong Baptist University EMAIL Jie Chen Hong Kong Baptist University EMAIL Junkun Jiang Hong Kong Baptist University EMAIL Jingyu Xiang Hong Kong Baptist University EMAIL |
| Pseudocode | Yes | Algorithm 1 Compositional Phase Diffusion on generating sequence composed by 2 semantic conditioned segments |
| Open Source Code | Yes | Codes are available at https://github.com/asdryau/Trans Phase. |
| Open Datasets | Yes | We use the BABEL-TEACH dataset [4, 12] for training and evaluation, as it provides annotated subsequence pairs essential for long-term motion generation [12, 13, 14], facilitating the learning of transitions between subsequences... Moreover, we follow PCMDM [14] and prior MDM [13] to transform the motion data into Human ML3D [6] format. |
| Dataset Splits | Yes | As a result, the training dataset contains 4370 subsequence pairs, while the testing dataset includes 1582 subsequence pairs. |
| Hardware Specification | No | The paper does not explicitly mention specific hardware details (GPU/CPU models, memory, etc.) used for running the experiments in the main text or Appendix A. |
| Software Dependencies | No | The paper mentions software components like "DDIMScheduler [3] FD is utilized" and "CLIP-Vi T-B/32 [23]" but does not provide specific version numbers for these or other key software libraries or programming languages. |
| Experiment Setup | Yes | We set the minimum and maximum lengths for each subsequence as 45 and 250 frames, respectively... Also, our models are designed based on phase latent size Q = 512, which serves as both the latent dimension for all diffusion modules and the number of periodic signals in ACT-PAE. For the diffusion step setting in SPDM and TPDM, DDIM [3] is utilized for 1000 training steps and 100 inference steps. |