Accurate and Steady Inertial Pose Estimation through Sequence Structure Learning and Modulation

Authors: Yinghao Wu, chaoran wang, Lu Yin, Shihui Guo, Yipeng Qin

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments across multiple benchmark datasets demonstrate the superiority of our approach against state-of-the-art methods and has the potential to advance the design of the transformer architecture for fixed-length sequences.
Researcher Affiliation Academia 1School of Informatics, Xiamen University, China 2School of Computer Science & Informatics, Cardiff University, UK
Pseudocode No The paper does not contain pseudocode or clearly labeled algorithm blocks. It describes the architecture and processes in text and diagrams.
Open Source Code No Yes, we will release the code/data later.
Open Datasets Yes We use the following datasets in our experiments, which can be divided into three categories: 1) Synthetic dataset: AMASS [32]. 2) Real datasets with SMPL [29] skeleton: DIP-IMU [18] and Total Capture [46]. 3) Real datasets with Xsens [41] skeleton: An Dy [33], CIP [37], and Emokine [9].
Dataset Splits No The paper mentions training and testing sets (e.g., "fine-tune it on the training set of DIP-IMU, then test it on the test set of DIP-IMU"), but does not explicitly define or specify a separate validation dataset split.
Hardware Specification Yes We implement our method using the Py Torch [40] framework on one NVIDIA Ge Force RTX 4090 GPU. ... We implement the live demo using a laptop equipped with an Intel Core i9-13900HX Processor CPU and an NVIDIA Ge Force RTX 4060 GPU.
Software Dependencies Yes We implement our method using the Py Torch [40] framework on one NVIDIA Ge Force RTX 4090 GPU. Py Torch version is 2.0.0, and CUDA version is 11.8.
Experiment Setup Yes During the training stage, we use the Adam W [30] optimizer to train our model with a batch size of 4096. The learning rate is initialized to 0.0001 and decayed by 0.99 per epoch.