Local-Global Feature Fusion for Enhancing 3D Human Pose Estimation

This work is based on the MotionAGFormer and P-STMO, and you can get more help there.

Environment

The code is conducted under the following environment:

• Ubuntu 20.04
• Python 3.9.16
• PyTorch 1.13.1
• CUDA 11.7

Dataset

The dataset setting follow the MotionAGFormer. Please refer to it to set up datasets (under ./data directory).

Evaluation

• Download our pretrained model from Google Drive;

Then run the command below (evaluate on 243 frames input):

• Human3.6M

python train.py --eval-only --checkpoint checkpoint --checkpoint-file 37.7h36m.pth.tr --config configs/h36m/LG3DPose.yaml

• MPI-INF-3DHP

python train.py --eval-only --checkpoint checkpoint --checkpoint-file 16.4mpi.pth.tr --config configs/mpi/LG3DPose.yaml

Training from scratch

Training our model with GPU:

• Human3.6M

python train.py --config configs/h36m/LG3DPose.yaml --use-wandb --wandb-name LG3DPose

• MPI-INF-3DHP

python train.py --config configs/mpi/LG3DPose.yaml --use-wandb --wandb-name LG3DPose-MPI

Acknowledgement

Thanks for the baselines, we construct the code based on them:

• MotionAGFormer
• P-STMO