Skip to content
Comics dataset for visual sentiment analysis
Branch: master
Clone or download
折栋宇 update
Latest commit 02a7bba Aug 12, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information. update Aug 12, 2019
example.png update Aug 12, 2019

Comics Dataset for Visual Sentiment Recognition

The Comics datset contains more than 10,000​ images belongs to 8​ categories. In specific, we use Amusement, Awe, Contentment, Excitement as positive sentiment, and Anger, Disgust, Fear, Sadness as negative sentiment. The images are collected from about seventy comics of various countries (e.g., America, Japan, China and France). About seventy comics are selected as candidates, e.g., Sponge Bob, Spiderman, The Avengers, One Piece, Slam Dunk, etc. Finally, a total of 10,281​ comic images are selected and roughly divided into Comic subset and Manga subset.

File Structure

The Comics dataset can be downloaded from Baidu or Google. The folders are arranged like this:

├── annotation
│   ├── test.txt
│   ├── train.txt
│   ├── trainval.txt
│   └── val.txt

Class Examples

Dataset Statistics

Class Amusement Awe Contentment Excitement Anger Disgust Fear Sad Total
Comic set 274 102 829 454 456 360 865 264 3604
Manga set 1220 327 1302 422 1128 388 1151 739 6677
All 1494 429 2131 876 1584 748 2116 1003 10281

The final version is slightly different from the paper.


If you find the Comics helpful, please cite it as

  title={Learning Discriminative Sentiment Representation from Strongly- and Weakly-Supervised CNNs},
  author={Dongyu She, Ming Sun, Jufeng Yang},
  journal={ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM)},

ATTN: This dataset is free for academic usage. For other purposes, please contact Dongyu She (

You can’t perform that action at this time.