Download the DHP19 dataset following this repo.
Run Generate_DHP19_pointcloud_all_mean.m to generate MeanLabel setting data. Or run Generate_DHP19_pointcloud_all_last.m for the LastLabel setting data.
You need to generate train data and test data seperately and set path for both. Following the DHP19, S1-S12 for training and S13-S17 for testing.
Note that you should change the data path corresponding to your own path in the two files above. You may get .h5 files at this step.
# For MeanLabel train/test
# Extract data and labels per frame from the h5 files, need to change path for train/test.
python extract_data_mean.py
python extract_labels_mean.py
python extract_3Dlabels.py
# Extract dict storing the event point numbers, which is used when training. Need to change path for train/test.
python extract_dict.py
# For LastLabel
# Extract data and labels per frame from the h5 files. Need to change path for train/test.
python extract_data_last.py
python extract_labels_last.py
Note that you should change the data path corresponding to your own path in the files above. You may get .npy files at this step.
For more detatils of MeanLabel and LastLabel, you can refer to our paper.
Your dataset folder will look like this:
├── DHP19EPC_dataset # Store test/train data
| ├─ test_LastLabel
| ├─ ├─data
| ├─ ├─├─S13_session1_mov1_cam0.h5
| ├─ ├─label
| ├─ ├─├─S13_session1_mov1_cam0_label.h5
| ├─ test_LastLabel_extract
| ├─ test_MeanLabel
| ├─ test_MeanLabel_extract
| ├─ train_LastLabel
| ├─ train_LastLabel_extract
| ├─ ├─data
| ├─ ├─├─S1_session1_mov1_cam0_frame0.npy
| ├─ ├─label
| ├─ ├─├─S1_session1_mov1_cam0_frame0_label.npy
| ├─ train_MeanLabel
| ├─ ├─data
| ├─ ├─├─S1_session1_mov1.h5
| ├─ ├─label
| ├─ ├─├─S1_session1_mov1_label.h5
| ├─ train_MeanLabel_extract
| ├─ ├─3Dlabel
| ├─ ├─├─S1_session1_mov1_frame0_3Dlabel.npy
| ├─ ├─data
| ├─ ├─├─S1_session1_mov1_frame0_cam0.npy
| ├─ ├─label
| ├─ ├─├─S1_session1_mov1_frame0_cam0_label.npy
| ├─ ├─Point_Num_Dict.npy
| ├─ ├─Point_Num_Dict_PerVideo.npy