Implementation of our CVPR 2024 paper "Loose Inertial Poser: Motion Capture with IMU-attached Loose-Wear Jacket". Including network weights, training and evaluation scripts.
Run train_SemoAE.py to train SeMo-AE.
Run train_poser.py to train pose estimation network.
Run evaluation.py to get ang Err and pos Err.
The real-world dataset is in the folder named LIP_Dataset, please notice that the original data frame rate is 60Hz, and it was downsampled to 30Hz in our implementation.
The original data frame rate is 60Hz, and it was downsampled to 30Hz which is a 50% reduction.