/V-DMGNN-GAN

Variational DMGNN GAN built upon DMGNN by Li et.al. Course project for EECE 571F, 2021WT2 at UBC.

Primary LanguagePythonMIT LicenseMIT

Variational DMGNN GAN

Variational seq2seq DMGNN-based GAN (V-DMGNN-GAN) model for human skeletal motion prediction. Course project for EECE 571F: Deep Learning with Structures at UBC, 2021 Winter Term 2.

Refer to here for the readme file from the original DMGNN paper.

Authors

Anushree Bannadabhavi*, Guanxiong Chen*, Yunpeng (Larry) Liu*, Kaitai (Alan) Tong*.

  • Authors listed by alphabetical order of last names. Equal contribution from all.

Instruction for training our GAN model

CMU dataset random masking

cd v-dmgnn-gan_cmu
python main.py prediction -c ../config/CMU/v-dmgnn-gan/train_random.yaml

ACCAD dataset lower body masking

cd v-dmgnn-gan_amass
python main.py prediction -c config/ACCAD/v-dmgnn-gan/train_lower-body.yaml

Instruction for evaluating our GAN model

CMU dataset random masking

cd v-dmgnn-gan_cmu
python main.py prediction -c ../config/CMU/v-dmgnn-gan/test_random.yaml

ACCAD dataset lower body masking

cd v-dmgnn-gan_amass
python main.py prediction -c config/ACCAD/v-dmgnn-gan/test_lower-body.yaml