Non-Local Latent Relation Distillation for Self-Adaptive 3D Human Pose Estimation

This repository is the official implementation of Non-Local Latent Relation Distillation for Self-Adaptive 3D Human Pose Estimation

Requirements

To install the requirements, please create a virtual environment and install the required packages. This codebase was created for and tested on Python 3.8:

python -m venv env
source env/bin/activate
pip install -r requirements.txt

Training

To train the model(s) in the paper, run the corresponding commands from the root directory:

Pose autoencoder

cd pose_autoencoder
python train.py

Motion autoencoder

cd motion_autoencoder
python train.py

Relation Transformer Networks

1. To train Motion Relation Transformer Networks:

cd relation_transformer
python train_motion_rule.py

2. To train Pose Relation Transformer Network:

cd relation_transformer
python train_pose_rule.py

Source image-to-latent model

cd image_to_latent_encoder
python train.py 

Target image-to-latent model. (Self-adaptation)

cd target_adaptation
python train.py

Evaluation

To evaluate the trained target model on the H3.6M test set, run the following commands:

cd evaluation
python eval.py