docs/pretrain.md

Pretrain

Data

AMASS

1. Please download data from the official website (SMPL+H).
2. We provide the preprocessing scripts as follows. Minor modifications might be necessary.
   • tools/compress_amass.py: downsample the frame rate
   • tools/preprocess_amass.py: render the mocap data and extract the 3D keypoints
   • tools/convert_amass.py: slice them to motion clips

Human 3.6M

Please refer to pose3d.md.

PoseTrack

Please download PoseTrack18 from MMPose (annotation files) and unzip to data/motion2d.

InstaVariety

1. Please download data from human_dynamics to data/motion2d.
2. Use tools/convert_insta.py to preprocess the 2D keypoints (need to specify name_action ).
3. Load all the processed .pkl files from step 2, concatenate them to motion_list, then run

import numpy as np
ids = []
for i, x in enumerate(motion_list):
    ids.append(np.ones(len(x))*i)
motion_all = np.concatenate(motion_list)
id_all = np.concatenate(ids)
np.save('data/motion2d/InstaVariety/motion_all.npy', motion_all)
np.save('data/motion2d/InstaVariety/id_all.npy', id_all)

You can also download the preprocessed 2D keypoints from here and unzip it to data/motion2d/.

The processed directory tree should look like this:

.
└── data/
    ├── motion3d/
    │   └── MB3D_f243s81/
    │       ├── AMASS
    │       └── H36M-SH
    ├── motion2d/
    │   ├── InstaVariety/
    │   │   ├── motion_all.npy
    │   │   └── id_all.npy
    │   └── posetrack18_annotations/
    │       ├── train
    │       └── ...
    └── ...

Train

python train.py \
--config configs/pretrain/MB_pretrain.yaml \
-c checkpoint/pretrain/MB_pretrain