search

EricGuo5513 / text-to-motion

Official implementation for "Generating Diverse and Natural 3D Human Motions from Texts (CVPR2022)."
MIT License
428 stars 38 forks source link

Generating results for the KIT ML dataset #2

Closed Rusali28 closed 2 years ago

Rusali28 commented 2 years ago

Hello, Can you kindly explain how can we get the sampling results for the KITML dataset? The instructions in the repo are for the HumanML3D

EricGuo5513 commented 2 years ago

Hi, sorry for KITML dataset, we only report the quantitative results as in "final_evaluation.py". To visualize the sampled results, you may need to modify the eval_comp_v6.py or gen_motion_script.py with loading the models for KIT-ML dataset and configurations as below:

elif opt.dataset_name == 'kit':
    opt.data_root = './dataset/KIT-ML'
    opt.motion_dir = pjoin(opt.data_root, 'new_joint_vecs')
    opt.text_dir = pjoin(opt.data_root, 'texts')
    opt.joints_num = 21
    radius = 240 * 8
    fps = 12.5
    dim_pose = 251
    opt.max_motion_length = 196
    kinematic_chain = paramUtil.kit_kinematic_chain