English | 简体中文

AffectNet Dataset Preprocessing Tool

This Python tool can help you organize the uncompressed AffectNet dataset into a structure that can be directly read by Pytorch's ImageFolder method. This tool can help you start using the AffectNet dataset for deep learning research more quickly.

How to Use

• Download and uncompress the AffectNet dataset(For most research, you only need to decompress the image compressed file Manually_Annotated.partX.rar in the Manually_Annotated folder and the corresponding label file Manually_Annotated_file_lists.zip).
• Place this tool in the root directory of the AffectNet dataset.
• Run the affectnet_preprocess.py file as

python affectnet_preprocess.py

• Or you can run the one-click processing script process.sh as:

bash process.sh

It will generate a folder AffectNet_class to store the processed dataset.

Notes

• Before running this tool, make sure you have uncompressed the AffectNet dataset.
• After processing, there should be train and test folders under the dataset folder, and the eight expressions in it are stored in folders 0-7.
• We also provide file structure reconstruction scripts reconstruction.sh in case an error occurs during processing and needs to be restarted. run with the following code：

bash reconstruction.sh

Contact Us

If you have any questions or issues while using this tool, please contact us.