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 [1].
ImDy: Human Inverse Dynamics from Imitated Observations
Authors: Xinpeng Liu, Junxuan Liang, Zili Lin, Haowen Hou, Yong-Lu Li, Cewu Lu
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on Im Dy and real-world data demonstrate the impressive competency of Im Dy S in human inverse dynamics and ground reaction force estimation. Moreover, the potential of Im Dy(-S) as a fundamental motion analysis tool is exhibited with downstream applications. The project page is https://foruck.github.io/Im Dy. ... Extensive experiments are conducted with analyses of the proposed data-driven methodology, demonstrating the feasibility of Im Dy S. |
| Researcher Affiliation | Academia | 1Shanghai Jiao Tong University, 2Shanghai Innovation Institute, 3Soochow University EMAIL, EMAIL EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methods and data acquisition pipeline but does not include any clearly labeled pseudocode or algorithm blocks. Figures 2, 3, and 16 illustrate system overviews and data flow, but are not formal pseudocode. |
| Open Source Code | No | The paper mentions a project page: "The project page is https://foruck.github.io/Im Dy.", but does not explicitly state that source code for the described methodology is available there or provide a direct link to a code repository. |
| Open Datasets | Yes | Im Dy contains over 150 hours of motion with joint torque and full-body ground reaction force data. ... We adopt AMASS (Mahmood et al., 2019) and KIT (Krebs et al., 2021) as two major data sources. ... Then Im Dy S is evaluated on Ground Link (Han et al., 2023), which contains real-world ground reaction force. Furthermore, we demonstrate the efficacy of Im Dy on the recent real-world human dynamics dataset Add Biomechanics (Werling et al., 2025). |
| Dataset Splits | Yes | We split Im Dy into a training set of 27,501 sequences and a test set of 3,055 sequences. ... We follow the train/test split in Addbiomechanics and report m PJE for the joint torque normalized by body weight. |
| Hardware Specification | Yes | All the data collection processes and experiments are conducted on a single NVIDIA RTX3090 GPU. |
| Software Dependencies | No | The paper mentions using specific software components like Isaac Gym and the Adam W optimizer, but does not provide specific version numbers for these or other libraries used for reproducibility. |
| Experiment Setup | Yes | The window size w is set as 2 to keep a shortterm motion modeling, which is proven helpful in Sec. 5.3. The encoder of Im Dy S is a threelayer transformer with a dimension of 64, ReLU activation, and Layer Norm. The loss weights are set as α1 = α3 = 0.01, α2 = α4 = α5 = 1 to maintain all terms at similar numerical scales for training stability. Im Dy S, the prior discriminator, and the FD model are all trained using the Adam W optimizer with a batch size of 2,400 for 140 epochs on Im Dy for the first stage. For the second stage, Im Dy S is further tuned on Add Biomechanics for only 10 epochs with the same hyperparameters. When generating negative samples for the prior discriminator, the two strategies are randomly adopted with a positive-negative ratio of 1:1. |