ShunliWang/CPR-Coach

Packages Check

The complete file list is shown below.

.
├── CPR_Dataset_S0.tar.gz.00            # 10 GB
├── CPR_Dataset_S0.tar.gz.01            # 10 GB
├── CPR_Dataset_S0.tar.gz.02            # 10 GB
├── CPR_Dataset_S0.tar.gz.03            # 10 GB
├── CPR_Dataset_S0.tar.gz.04            # 10 GB
├── CPR_Dataset_S0.tar.gz.05            # 10 GB
├── CPR_Dataset_S0.tar.gz.06            # 10 GB
├── CPR_Dataset_S0.tar.gz.07            # 5.4 GB
├── CPR_Dataset_S1.tar.gz.00            # 10 GB
├── CPR_Dataset_S1.tar.gz.01            # 10 GB
├── CPR_Dataset_S1.tar.gz.02            # 10 GB
├── CPR_Dataset_S1.tar.gz.03            # 10 GB
├── CPR_Dataset_S1.tar.gz.04            # 10 GB
├── CPR_Dataset_S1.tar.gz.05            # 10 GB
├── CPR_Dataset_S1.tar.gz.06            # 10 GB
├── CPR_Dataset_S1.tar.gz.07            # 4.9 GB
├── CPR_Double_Dataset_S0.tar.gz.00     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.01     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.02     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.03     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.04     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.05     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.06     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.07     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.08     # 10 GB
├── CPR_Double_Dataset_S0.tar.gz.09     # 2.7 GB
├── CPR_Double_Dataset_S1.tar.gz.00     # 10 GB
├── CPR_Double_Dataset_S1.tar.gz.01     # 10 GB
├── CPR_Double_Dataset_S1.tar.gz.02     # 10 GB
├── CPR_Double_Dataset_S1.tar.gz.03     # 10 GB
├── CPR_Double_Dataset_S1.tar.gz.04     # 10 GB
├── CPR_Double_Dataset_S1.tar.gz.05     # 10 GB
├── CPR_Double_Dataset_S1.tar.gz.06     # 10 GB
├── CPR_Double_Dataset_S1.tar.gz.07     # 10 GB
├── CPR_Double_Dataset_S1.tar.gz.08     # 9.6 GB
├── SupC14_Dataset.tar.gz               # 9.1 GB
├── SupDC59_Dataset.tar.gz.00           # 10 GB
├── SupDC59_Dataset.tar.gz.01           # 10 GB
├── SupDC59_Dataset.tar.gz.02           # 10 GB
├── SupDC59_Dataset.tar.gz.03           # 8.9 GB
├── SupMC15_Dataset.tar.gz.00           # 10 GB
├── SupMC15_Dataset.tar.gz.01           # 10 GB
├── SupMC15_Dataset.tar.gz.02           # 6.0 GB
├── Keypoints.tar.gz                    # 2.5 GB
└── ann.tar.gz                          # 128KB

Run the Preprocessing Script

We have written the preprocessing code of the CPR-Coach dataset. The preprocessing includs a series of merge, decompression and deletion operations. The whole process may take 20-30 minutes. Since the parallel processing is adopted, program interruption should be avoided. After running the process.sh file, you can get the entire dataset.

sh ./process.sh

Dataset Check

After the preprocessing, make sure that the structure of the dataset is same with the follow. The total size of the dataset should be 451 GB.

.
├── CPR_Dataset_S0
│   ├── C00_S0
│   ├── C01_S0
│   ├── C02_S0
│   └── ...
├── CPR_Dataset_S1
│   ├── C00_S1
│   ├── C01_S1
│   ├── C02_S1
│   └── ...
├── CPR_Double_Dataset_S0
│   ├── DC00_S0
│   ├── DC01_S0
│   ├── DC02_S0
│   └── ...
├── CPR_Double_Dataset_S1
│   ├── DC00_S1
│   ├── DC01_S1
│   ├── DC02_S1
│   └── ...
├── SupC14_Dataset
│   ├── SupC00
│   ├── SupC01
│   ├── SupC02
│   └── ...
├── SupDC59_Dataset
│   ├── SupDC00
│   ├── SupDC01
│   ├── SupDC02
│   └── ...
├── SupMC15_Dataset
│   ├── SupMC00
│   ├── SupMC01
│   ├── SupMC02
│   └── ...
├── Keypoints
│   ├── all_errors_keypoints.pkl
│   ├── double_errors_keypoints.pkl
│   ├── test_keypoints.pkl
│   └── train_keypoints.pkl
├── ann
│   ├── ActionList.txt
│   ├── all_errors_testlist.txt # Contains All paired-, triple- and quad-errors Samples
│   ├── double_errors_testlist.txt # Contains paired-errors Samples
│   ├── testlist.txt # Used for Single-class Testing
│   ├── testlist_of.txt # Used for Single-class Testing (Optical Flow)
│   ├── trainlist.txt # Used for Single-class Training
│   ├── trainlist_of.txt # Used for Single-class Training (Optical Flow)
│   └── triquad_errors_testlist.txt # Contains triple- and quad-errors Samples
└── process.sh

Dataset Structure Description

In the initial version of the dataset, we only collected videos of a single person dressed in two different clothes. And only 59 types of double errors composite actions were explored. These data exist in CPR_Dataset_S0, CPR_Dataset_S1, CPR_Double_Dataset_S0 and CPR_Double_Dataset_S1.

We put the initial version of this work into CVPR-2023, and the reviewers gave us valuable suggestions: further enrich the types of samples and composite errors.

Therefore, we subsequently improved this dataset and recruited several volunteers to conduct supplementary data collection. These data are stored in SupC14_Dataset, SupDC59_Dataset and SupMC15_Dataset. We divided these samples in the ann directory.