Data Preparation

Easy Version

We have successfully fine-tuned our VideoMAE on AVA v2.2 with this codebase.

The pre-processing of AVA v2.2 can be summarized into 2 steps:
1. Download the processed dataset from Google Drive or Baidu NetDisk (code q5v5).
2. run following commands to unzip the file and create a symbolic link to the extracted files.
```
tar zxvf AVA_compress.tar.gz -C /your/path/
cd /path/to/VideoMAE-Action-Detection/
ln -s /your/path/AVA data/AVA
```

Step-by-step Version

Please follow the instructions in AlphAction for step-by-step data preparation.