search

jyf588 / transformer-inertial-poser

Python implementation accompanying the Transformer Inertial Poser paper at SIGGRAPH Asia 2022
https://arxiv.org/abs/2203.15720
Other
66 stars 10 forks source link

Motion visualization #5

Closed Jaceyxy closed 1 year ago

Jaceyxy commented 1 year ago

Thanks for such good work， I wonder if it is possible to visualize the results of training, for example, for the cmu data set

jyf588 commented 1 year ago

Just in case other people are curious in the future, it's pretty much the same as visualizing/testing on the eval datasets. For example, one could just run python offline_testing_simple.py --name_contains "40_40_03" --ours_path_name_kin output/your_own_model.pt --with_acc_sum --test_len 30000 --compare_gt --seed 42 --five_sbp --render

The only differences here are that

  1. use name_contains to specify the preprocessed motions you want to run the model on. For example, "40_40_03" will run the model on all motions containing the substring "40_40_03". (there is only one motion with such name and it's part of the CMU dataset, assuming you have run sh ./data-gen-new-scripts.bash)

  2. use the --render option to enable pyBullet rendering.

Jaceyxy commented 1 year ago

Thanks very much！