This is the offical Pytorch implementation of PersPose, which estimates the 3D positions of joints from individual images. Below is the overall framework.

This code is tested on Ubuntu 20.04 with Pytorch 1.13.1 and Python 3.9.17.

conda create -n perspose python=3.9.17
conda activate perspose
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
pip install -r requirements.txt
wandb login # paste your wandb API key when prompted.

1. Download SMPL model files ( basicmodel_f_lbs_10_207_0_v1.1.0.pkl, basicmodel_m_lbs_10_207_0_v1.1.0.pkl, and basicmodel_neutral_lbs_10_207_0_v1.1.0.pkl) to ./data/smpl_file/. Please rename them as SMPL_FEMALE.pkl, SMPL_MALE.pkl, and SMPL_NEUTRAL.pkl, respectively.

2. Download SMPL related files (J_regressor_h36m_correct.npy, segm_per_v_overlap_from_surreal.pkl) to ./data/smpl_file/.

3. To train or evaluate, please prepare and link datasets to ./data/. Also, download processed annotation files for all datasets. (SMPL parameters for Human3.6M are from Mosh.)

   Dataset	Image path example
   3DPW	./data/3dpw/imageFiles/courtyard_arguing_00/image_00000.jpg
   Human3.6M	./data/h36m/images/s_01_act_02_subact_01_ca_01/s_01_act_02_subact_01_ca_01_000001.jpg
   MPIINF-3DHP	./data/3dhp/mpi_inf_3dhp_train_set/S1/Seq1/images/S1_Seq1_V0/img_S1_Seq1_V0_000001.jpg
   COCO	./data/coco/train2017/000000155582.jpg
   BEDLAM (Optional)	./data/bedlam/data/20221019_3_250_highbmihand/png/seq_000000/seq_000000_0000.png

4. To train, please download the pretrained HRNet checkpoint, pose_hrnet_w48_256x192.pth, to ./data/ckpt/.

To inference, you need to prepare python environment as detailed in the Installation section above. Also, you need to prepare SMPL model files and SMPL related files as detailed in the Data Preparing section above. Moreover, you need to download a model weight file, e.g. ckpt_w_be_3dhp_35epochs.pth, from here.

Next, prepare input images and the paired focal length. The photos taken by cameras or some phones have a 35 mm equivalent focal length attribute in their EXIF metadata. If not, you can first measure the horizontal FOV of your camera mannually, and then the 35 mm equivalent focal length can be calculated as: focal_length_35 = 36/2 / tan(FOV_Horizontal/2). (Width of 35 mm film: 36mm.) Then, please add it to photos' EXIF, or rename your input images as {base_filename}_{focal_length_35}_f35.jpg, e.g. IMG_26_f35.jpg.

Finally, run the following command to inference, taking the images in directory ./demo/to_inference as inputs, and there will be (PR) centered image crops and visualization of inference results in the folder ./demo/to_inference_res.

CUDA_VISIBLE_DEVICES=0 python ./src_PersPose/trainer.py --cfg ./src_PersPose/configs/config_w_be_3dhp.yaml --pretrained_ckpt ./data/ckpt/PersPose/ckpt_w_be_3dhp_35epochs.pth --inference ./demo/to_inference

Every experiment is defined by config files. Configs of the experiments in the paper can be found in the ./src_PersPose/configs/ directory.

Run the following command to train on a single GPU without BEDLAM dataset.

CUDA_VISIBLE_DEVICES=0 python ./src_PersPose/trainer.py --cfg ./src_PersPose/configs/config_3dpw.yaml --wandb 3dpw_wo_be

Or run the following command to train on multiple(4) GPUs without BEDLAM dataset.

CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --rdzv_backend=c10d --rdzv_endpoint=localhost:0 --nnodes=1 --nproc_per_node=4 ./src_PersPose/trainer.py --cfg ./src_PersPose/configs/config_3dpw.yaml --wandb 3dpw_wo_be

Run the following command to evaluate the provided checkpoints (model weights) on a single GPU.

CUDA_VISIBLE_DEVICES=0 python ./src_PersPose/trainer.py --cfg ./src_PersPose/configs/config_3dpw.yaml --test --pretrained_ckpt ./data/ckpt/PersPose/ckpt_3dpw_16epochs.pth

Or run the following command to evaluate on multiple(4) GPUs.

CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --rdzv_backend=c10d --rdzv_endpoint=localhost:0 --nnodes=1 --nproc_per_node=4 ./src_PersPose/trainer.py --cfg ./src_PersPose/configs/config_3dpw.yaml --test --pretrained_ckpt ./data/ckpt/PersPose/ckpt_3dpw_16epochs.pth

Cite as below if you find this repository helpful:

@InProceedings{PersPose_2025_ICCV,
    title     = {PersPose: 3D Human Pose Estimation with Perspective Encoding and Perspective Rotation},
    author    = {Hao, Xiaoyang and Li, Han},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    year      = {2025},
}

This repo is built on the excellent works HybrIK and VirtualMarker.