A educational project (e.g., using 3D pose to control 3D carton) for learning 3D vision (application of human mesh recovery) or imitation learning
2021/06/08: Update the requirement for your code submission on github
2021/05/20: Update the saving logic of mapping results(.pkl). In previous version, pkl will not be overwritten, so if you try mapping multiple times, the pkl will stay as it was in previous mapping. This problem is fixed now, pls ensure that your results/ is the latest version
2021/05/17: Add quick start instructions to help you dive into this project & Use custom obj loader instead of open3d
- code only tested on linux system (ubuntu 16.04)
- open3d==0.11.0(if you want to try online models, use open3d 0.10.0 along with your own video recording tool to save visualization)
- tqdm
- opencv-python
pip install -r requirements.txt (anaconda recommended, python3)
- For maya download & install: Tutorial1, Tutorial2 Tutoria3
- For SMPL, refer to webpage or popular implementation Minimal-IK
- For online model matching, refer to OnlineMatchTutorial
transfer.py: the main mapping filevis.py: visualize the mapping sequence of the corresponding meshpose_samples: some samples of SMPL model for one frameobj_seq_id: some samples of SMPL model for temporal sequence
Run vis.py (to get more clear visualization, press ctrl + 1 / ctrl + 2 / ctrl + 3 ...)
The code is under Apache-2.0 License.
Please read course project requirements.