I. Installation

conda create --name openmmlab python=3.8 -y
conda activate openmmlab
pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113
pip install -U openmim
mim install mmengine
mim install mmcv

II. Data Preparation

1. Build Dataset from Scratch

Create the Entire Dataset

• Run the following commands:

  bash create_gt.sh
  python create_data.py

  This will generate a file named all.csv.

Create K-Folds

To split the data into training and testing sets similar to the utvm-vtcc team, use:

python create_kfolds.py

2. Download from Drive

Alternatively, you can download the dataset directly. Download 0.zip and 1-15.zip from this Google Drive folder and unzip them.

Data Structure

The dataset folder should have the following structure:

dataset
├── 0
├── 1
├── 2
├── 3
├── 4
├── 5
├── 6
├── 7
├── 8
├── 9
├── 10
├── 11
├── 12
├── 13
├── 14
└── 15

III. Train the Model

Make sure to update the data_path variable to point to your dataset directory. Then run the training script:

bash run.sh

Log Wandb

IV. Validation

1. Navigate to the infer directory:

   cd infer

2. Download the final_models.zip file from this Google Drive folder and unzip it.

3. The folder structure for the models should look like this:

   final_models
   ├── tf_efficientnetv2_l_in21k imgs 256 bs 16
   │   ├── 2023_09_14_09.48
   │   └── 2023_09_14_17.18
   ├── tf_efficientnetv2_m_in21k imgs 256 bs 48
   │   ├── 2023_09_09_16.11
   │   └── 2023_09_09_19.32
   └── tf_efficientnetv2_s_in21k imgs 256 bs 48
       ├── 2023_09_13_17.46
       └── 2023_09_13_20.17
   

4. Run the final inference:

   python final_inference.py

Compare Different Methods

You can also experiment with training other algorithms like:

• SlowFast
• C3D
• TANet
• TSN

For more details, refer to the "mmaction2/readme.md" file.

References

• Docker File
• Dataset Creation by Divedeeper - Team 83
• Pose Extraction using mmaction2