DSAG

The official pytorch implementation of DSAG: A Scalable Deep Framework for Action-Conditioned Multi-Actor Full Body Motion Synthesis, WACV 2023. Please visit our webpage for more details.

Table of contents:

1. Description
2. How to use

Description

We introduce DSAG, a controllable deep neural framework for action-conditioned generation of full body multi-actor variable duration actions. To compensate for incompletely detailed finger joints in existing large-scale datasets, we introduce full body dataset variants with detailed finger joints. To overcome shortcomings in existing generative approaches, we introduce dedicated representations for encoding finger joints. We also introduce novel spatiotemporal transformation blocks with multi-head self attention and specialized temporal processing. The design choices enable generations for a large range in body joint counts (24 - 52), frame rates (13 - 50), global body movement (in-place, locomotion) and action categories (12 - 120), across multiple datasets (NTU-120, HumanAct12, UESTC, Human3.6M). Our experimental results demonstrate DSAG's significant improvements over state-of-the-art, its suitability for action-conditioned generation at scale.

How to use

Requirements

$ pip install -r requirements.txt 

Data Preprocessing

NTU-Xpose dataset

We derive NTU-Xpose dataset for our experiments on the NTU dataset. To extract NTU-Xpose dataset, ExPose pose estimation used on the NTU RGBD-120 dataset. But ExPose extracts only the local pose, i.e. human pose, with respect to the root joint. For Extracting the position of the human pose at the global trajectory, we use the NTU-120 Kinect dataset. The whole process of data preprocessing is demonstrated below.

Step 1: Extract NTU RGBD-120 skeleton using ExPose pose estimation, and place it in ./dataset/smpl directory. Then run the following,

$ mkdir -p dataset/smpl/
$ python ./Data_preprocessing/data_preprosessing_2_persons_init_75.py 

Step 2 Run the following for extracting and mapping root trajectory from 'NTU-120' dataset.

$ python ./Data_preprocessing/map_2_persons.py 

step 3 In this step we apply optimization based temporal consistency,

$ python ./Data_preprocessing/preprocess_finger_motion_2_person.py

Training

You can skip this step if you want to use our pretrained model. To train DSAG from scratch, run pose.py. Details of DSAG model can be found in checkpoints/. For training DSAG run the following,

$ python pose.py

Visualization

If you want to play with pretrained DSAG model please see visualization.ipynb or the cloab notebook

Evaluation

Implementation of evaluation metrics can be found in the metrics directory.

Bibtex

Please cite our paper if you end up using it for your own research:

@InProceedings{DSAG,
title = {DSAG: A Scalable Deep Framework for Action-Conditioned Multi-Actor Full Body  Motion Synthesis},
author = {Debtanu Gupta, Shubh Maheshwari, Sai Shashank Kalakonda, Manasvi Vaidyula, Ravi Kiran Sarvadevabhatla},
booktitle = {WACV},
year = {2023},
}