This repository contains code for "There and Back Again: 3D Mesh Sequence Generation from Text Using Back-Translation".
Before using this code, please install the dependencies inside environment.yml.
Additionally, you will have to install the dependencies and download the models for SMPLify-X: https://github.com/vchoutas/smplify-x
- unip the model weights in demo/eval_model/sign_sample_model_200_r0t1_all_256_1024_2D_pose_hands
- unzip the test and dev archive in demo/TextData and put the files in the TextData folder
- unzip the test archive under demo/Data
- change the paths inside demo/config.yaml
- call run_demo.py with the following arguments:
--config /path/to/Text2Mesh/Pose2Mesh/smplifyx/cfg_files/fit_smplx.yaml --data_folder /path/to/Text2Mesh/demo/Data --output_folder /path/to/Text2Mesh/demo/Data --visualize=False --model_folder /path/to/Text2Mesh/Pose2Mesh/smplifyx/smplx/models/smplx --vposer_ckpt /path/to/Text2Mesh/Pose2Mesh/smplifyxs/smplifyx/V02_05 --part_segm_fn smplx_parts_segm.pkl
Full evaluation and training code will be added upon acceptance of the manuscript. The selected demo sequences include success and failure cases to give a fair representation.
This repo contains code adapted and extended from:
https://github.com/vchoutas/smplify-x
Same licences apply.