Deep Non-Rigid Structure-from-Motion Revisited: Canonicalization and Sequence Modeling

Authors: Hui Deng, Jiawei Shi, Zhen Qin, Yiran Zhong, Yuchao Dai

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The effectiveness of our method is verified by testing the sequence-to-sequence deep NRSf M pipeline with corresponding regularization modules on several commonly used datasets. ... We conducted experiments on a substantial amount of data, and the results confirm the validity of our method in most scenarios. ... We commence this section with a thorough depiction of the dataset employed as well as the metrics. Following this, the quantitative outcomes will be presented, and ultimately, the ablation experiments will be conducted, accompanied by a performance discussion.
Researcher Affiliation Collaboration 1 School of Electronics and Information, Northwestern Polytechnical University 2 Tap Tap, 3 Open NLPLab EMAIL, EMAIL, EMAIL
Pseudocode No The paper describes methods using mathematical equations and textual explanations, but it does not contain any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement of code release, a link to a repository, or a clear indication that code is available in supplementary materials for the methodology described.
Open Datasets Yes Human3.6M. This classic dataset contains a large number of human motion sequences annotated with 3D ground truth which are extracted by motion capture systems (Ionescu et al. 2014). ... Inter Hand2.6M. This dataset contains a large number of highly deformed hand poses (Moon et al. 2020). ... 3DPW. 3DPW(von Marcard et al. 2018) is a dataset commonly used for human pose estimation... CMU MOCAP. ... preprocess the CMU MOCAP datasest 1. We follow the splitting protocol of (Kong and Lucey 2021)...
Dataset Splits Yes CMU MOCAP. For a fair comparison with existing methods, we follow the setup of (Zeng et al. 2021) to preprocess the CMU MOCAP datasest 1. We follow the splitting protocol of (Kong and Lucey 2021), where 1/5 action sequences are used for testing, and the rest are used for training.
Hardware Specification Yes the sequence length is 32, the batch is 256, and 4 Nvidia RTX 3080Ti are used for training.
Software Dependencies No The paper mentions 'gated linear unit(Shazeer 2020)' and 'gated Toeplitz unit (Gtu) as (Qin et al. 2023)', which are architectural components, but it does not list specific software libraries or frameworks with version numbers (e.g., 'PyTorch 1.9', 'Python 3.8') that would allow replication of the experimental environment.
Experiment Setup Yes The experimental settings are, GPA convergence threshold is 1e 8, and the maximum number of iterations is 100. The weight of the loss function is α = 9, β = 0.1, the sequence length is 32, the batch is 256, and 4 Nvidia RTX 3080Ti are used for training.