A cross-feature interaction network for 3D human pose estimation

This is the official implementation for "A cross-feature interaction network for 3D human pose estimation (Pattern Recognition Letters 2025)" on PyTorch platform.

Dependencies

Make sure you have the following dependencies installed:

• Ubuntu 20.04
• CUDA 11.2
• Python 3.7.13
• PyTorch 1.8.1
• Matplotlib=3.1.0

Dataset

Our model is evaluated on Human3.6M and MPI-INF-3DHP.

Human3.6M

We set up the Human3.6M dataset in the same way as VideoPose3D.

MPI-INF-3DHP

We set up the MPI-INF-3DHP dataset in the same way as P-STMO.

Training from scratch

For training the model on Human3.6m using 2D keypoints obtained by CPN, please run:

python run.py -k cpn_ft_h36m_dbb --train --batch_size 512 --epoch 20

For training the model on Human3.6m using ground-truth 2D keypoints, please run:

python run.py -k gt --train --batch_size 256 --epoch 20

Evaluating our models

You can download our pre-trained models from Google Drive. Put model_cfi_gt.pth, model_cfi_cpn.pth in the ./checkpoint directory. Both of the models are trained on Human3.6M dataset.

To evaluate the model trained on Human3.6m using 2D keypoints obtained by CPN, run:

python run.py -k cpn_ft_h36m_dbb --evaluate --previous_dir checkpoint/model_cfi_cpn.pth

To evaluate the model trained on Human3.6m using ground-truth 2D keypoints, run:

python run.py -k gt --evaluate --previous_dir checkpoint/model_cfi_gt.pth

To evaluate the model on the test set of MPI-INF-3DHP, run:

python run.py --dataset 3dhp -k cpn_ft_h36m_dbb --evaluate --previous_dir checkpoint/model_cfi_cpn.pth

Acknowledgement

Our code refers to the following repositories.

• VideoPose3D
• P-STMO
• MHFormer

We thank the authors for releasing their codes. If you use our code, please consider citing our paper as well.