Regular Splitting Graph Network for 3D Human Pose Estimation (RS-Net)[IEEE Transactions on Image Processing'2023]

The PyTorch implementation for RS-Net.

Qualitative and quantitative results

Method	MPJPE(mm)	PA-MPJPE(mm)
SemGCN	57.6	-
High-order GCN	55.6	43.7
HOIF-Net	54.8	42.9
Weight Unsharing	52.4	41.2
ModulatedGCN	49.4	39.1
Ours	47.0	38.6

Dependencies

Make sure you have the following dependencies installed:

• PyTorch >= 1.7.0
• NumPy
• Matplotlib
• FFmpeg (if you want to export MP4 videos)
• ImageMagick (if you want to export GIFs)

You can create the environment:

conda create -n rsnet python=3.8
conda activate rsnet
pip install -r requirements.txt
pip install torch==1.7.0+cu110 torchvision==0.8.1+cu110 torchaudio==0.7.0 -f https://download.pytorch.org/whl/torch_stable.html

Dataset

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

Human3.6M & MPI-INF-3DHP

We set up the Human3.6M & MPI-INF-3DHP dataset in the same way as PoseAug. Please refer to DATASETS.md for the preparation of the dataset files & put them in ./dataset directory.

Evaluating our models

You can download our pre-trained models from here. Put them in the ./checkpoint directory.

Human 3.6M

To evaluate our pre-trained model using the detected 2D keypoints (HR-Net) with pose refinement, please run:

python main_graph.py -k hr --post_refine --rsnet_reload 1 --post_refine_reload 1 --save_out_type post --show_protocol2 --previous_dir './checkpoint/HR-Net' --rsnet_model model_rsnet_2_eva_post_4704.pth --post_refine_model model_post_refine_2_eva_post_4704.pth --nepoch 2 -z 96 --batchSize 512

To evaluate our pre-trained model using ground truth 2D keypoints without pose refinement, please run:

python main_graph.py -k gt --post_refine --rsnet_reload 1 --show_protocol2 --previous_dir './checkpoint/GT' --rsnet_model model_rsnet_5_eva_xyz_3728' --nepoch 2 -z 64 --batchSize 128

Training from scratch

Human 3.6M

To train our model using the detected 2D keypoints (HR-Net) with pose refinement, please run:

python main_graph.py -k hr --pro_train 1 --save_model 1  --save_dir './checkpoint' --show_protocol2  --post_refine --save_out_type post -z 96 --batchSize 512 --nepoch 31

To evaluate our model using the detected 2D keypoints (HR-Net) with pose refinement, please run:

python main_graph.py -k hr --post_refine --rsnet_reload 1 --post_refine_reload 1 --save_out_type post --show_protocol2 --previous_dir './checkpoint/HR-Net' --rsnet_model '[model_rsnet]' --post_refine_model '[model_post_refine]' --nepoch 2 -z 96 --batchSize 512

To train our model on the ground truth 2D keypoints without pose refinement, please run:

python main_graph.py -k gt  --pro_train 1 --save_model 1  --save_dir './checkpoint/GT' --show_protocol2  -z 64 --batchSize 128 --nepoch 31 --learning_rate 1e-3 --large_decay_epoch 5 --lr_decay .95

To evaluate our model using ground truth 2D keypoints without pose refinement, please run:

python main_graph.py -k gt --rsnet_reload 1 --show_protocol2 --previous_dir './checkpoint/GT' --rsnet_model '[model_rsnet]' --nepoch 2 -z 64 --batchSize 128

Acknowledgement

Our code refers to the following repositories.

• ModulatedGCN
• HOIF-Net
• PoseAug
• MHFormer

We thank the authors for releasing their codes.