SINC: Spatial Composition of 3D Human Motions for Simultaneous Action Generation

Nikos Athanasiou* · Mathis Petrovich* · Michael J. Black · Gül Varol

ICCV 2023

Official PyTorch implementation of the paper "SINC: Spatial Composition of 3D Human Motions for Simultaneous Action Generation"

Features

This implementation:

• Instruction on how to prepare the datasets used in the experiments.
• The training code:
  • For SINC method
  • For the baselines
  • For the ablations done in the paper
• Standalone script to compose different motions from AMASS automatically and create synthetic data from existing motions

Environment & Basic Setup

Details

SINC has been implemented and tested on Ubuntu 20.04 with python >= 3.10.

Clone the repo:

git clone https://github.com/athn-nik/sinc.git

After it do this to install DistillBERT:

cd deps/
git lfs install
git clone https://huggingface.co/distilbert-base-uncased
cd ..

Install the requirements using virtualenv :

# pip
source scripts/install.sh

You can do something equivalent with conda as well.

Data & Training

‼️⚠️ You can directly download the data from this link and use them!

Details

There is no need to do this step if you have followed the instructions and have done it for TEACH. Just use the ones from TEACH.

Step 1: Data Setup

Download the data from AMASS website. Then, run this command to extract the amass sequences that are annotated in babel:

python scripts/process_amass.py --input-path /path/to/data --output-path path/of/choice/default_is_/babel/babel-smplh-30fps-male --use-betas --gender male

Download the data from TEACH website, after signing in. The data SINC was trained was a processed version of BABEL. Hence, we provide them directly to your via our website, where you will also find more relevant details. Finally, download the male SMPLH male body model from the SMPLX website. Specifically the AMASS version of the SMPLH model. Then, follow the instructions here to extract the smplh model in pickle format.

The run this script and change your paths accordingly inside it extract the different babel splits from amass:

python scripts/amass_splits_babel.py

Then create a directory named data and put the babel data and the processed amass data in. You should end up with a data folder with the structure like this:

data
|-- amass
|  `-- your-processed-amass-data 
|
|-- babel
|   `-- babel-teach
|       `...
|   `-- babel-smplh-30fps-male 
|       `...
|
|-- smpl_models
|   `-- smplh
|       `--SMPLH_MALE.pkl

Be careful not to push any data! Then you should softlink inside this repo. To softlink your data, do:

ln -s /path/to/data

You can do the same for your experiments:

ln -s /path/to/logs experiments

Then you can use this directory for your experiments.

Step 2 (a): Training

To start training after activating your environment. Do:

python train.py experiment=baseline logger=none

Explore configs/train.yaml to change some basic things like where you want your output stored, which data you want to choose if you want to do a small experiment on a subset of the data etc. You can disable the text augmentations and using single_text_desc: false in the model configuration file. You can check the train.yaml for the main configuration and this file will point you to the rest of the configs (eg. model refers to a config found in the folder configs/model etc.).

Step 2 (b): Training MLD

Prior to running this code for MLD please create and activate an environment according to their repo. Please do the 1. Conda Environment and 2. Dependencies out of the steps in their repo.

python train.py experiment=some_name run_id=mld-synth0.5-4gpu model=mld data.synthetic=true data.proportion_synthetic=0.5 data.dtype=seg+seq+spatial_pairs machine.batch_size=16 model.optim.lr=1e-4 logger=wandb sampler.max_len=150

AMASS Compositions

Details

Given that you have downloaded and processed the data, you can create spatial compositions from gropundtruth motions of BABEL subset from AMASS using a standalone script:

python compose_motions.py

Evaluation

Details

After training, to sample and evaluate a model which has been stored in a folder /path/to/experiment

python sample.py folder=/path/to/experiment/ ckpt_name=699 set=small

python eval.py folder=/path/to/experiment/ ckpt_name=699 set=small

• You can change the jointstype for the sampling script to output and save rotations and translation by setting joinstype=rots.
• By setting the set=full you will obtain the results on the full BABEL validation set.

You can calculate the TEMOS score using:

python sample_eval_latent.py folder=/is/cluster/fast/nathanasiou/logs/space/single-text-baselines/rs_only/babel-amass/ ckpt_name=699 set=small

or for model trained using MLD:

python mld_temos.py folder=/is/cluster/fast/nathanasiou/logs/sinc/sinc-arxiv/mld-wo-synth/babel-amass ckpt_name=399 set=small

Citation

@inproceedings{SINC:ICCV:2022,
  title={{SINC}: Spatial Composition of {3D} Human Motions for Simultaneous Action Generation},
  author={Athanasiou, Nikos and Petrovich, Mathis and Black, Michael J. and Varol, G\"{u}l },
  booktitle = {ICCV},
  year = {2023}
}

License

This code is available for non-commercial scientific research purposes as defined in the LICENSE file. By downloading and using this code you agree to the terms in the LICENSE. Third-party datasets and software are subject to their respective licenses.

References

Many part of this code were based on the official implementation of TEMOS.

Contact

This code repository was implemented by Nikos Athanasiou and Mathis Petrovich.

Give a ⭐ if you like.

For commercial licensing (and all related questions for business applications), please contact ps-licensing@tue.mpg.de.