We provide some pretrained models to show the performance of the proposed method.

You can find them in data/.

• Models:

  • generator_x_4GRU.pkl
  • generator_y_4GRU.pkl
  • generator_z_4GRU.pkl
  • generator_v_4GRU.pkl

• Directions_1.npy and Directions_2.npy are the pre-processed test files corresponding to the S5 file in the Human 3.6M dataset.

Change the file read path in prediction_model.py to generate the predicted sequence for the specified action and calculate the MPJPE. The generated GT_X.npy is the Ground Truth, and vis_X.npy is the generated prediction sequence.

Change the file read path and the save path of the generated GIF image in vis_modle.py to generate the visualization results.

1. We train on action class X and test on class X.
2. Following existing works ( Learning dynamic relationships for 3d human motion prediction et al.), we use 17 joints to represent a skeleton.