Click to view and download the K-drama Facial Emotion Recognition Dataset User Agreement

This dataset is used in the following conference paper:

Heeryon Cho, Woo Kyu Kang, Younsoo Park, Sungeu Chae, and Seong-joon Kim, "Multi-label Facial Emotion Recognition Using Korean Drama Video Clips," in Proceedings of the 2022 IEEE International Conference on Big Data and Smart Computing (BigComp-22), IEEE, January 17-20, 2022, Daegu, South Korea.

To use this dataset for non-commercial research, please fill in and sign the User Agreement (PDF file) attached to this post, and send the scanned or photographed form to heeryon@cau.ac.kr. Once we receive your signed agreement, we will let you know how to download the dataset via a reply email.

!!! Please use this dataset for non-commercial research only. Any other use is strictly prohibited. !!!

This dataset contains 38,817 colored cropped face images tagged with 23 emotion labels. If we include 'neutral' emotion, then the total number of emotion labels is 24. The following lists the 24 emotion labels.

[affectionate, amused, angry, anxious, ashamed, bored, depressed, disappointed, disgusted, fearful, fluttered, fulfilled, guilty, happy, jealous, moved, passionate, peaceful, sad, smitten, surprised, sympathetic, upset, neutral (24)]

This dataset includes 19,800 images tagged with one label, 18,134 images tagged with two labels (i.e., multi-label), 875 images tagged with three labels, and 8 images tagged with four labels. 

While existing facial emotion recognition dataset predominantly captures Western (i.e., non-Asian) faces, our dataset covers various Korean (Asian) faces with wide age range, making this a valuable Asian multi-label facial emotion recognition dataset. 

We created three different kinds of datasets using the 38,817 images and labels.

• 6 basic emotion classification dataset (6 emotions)
• 22 emotion classification dataset (22 emotions)
• multi-label classification dataset (24 emotions)

All three datasets are split into train/validation/test data. All datasets are provided in tfrecord format.

You can download the experiment codes used in the conference paper at the first author's GitHub repository (https://github.com/heeryoncho/facial_emotion_recognition_using_K-drama_dataset).