Web Image Dataset for Event Recognition (WIDER)

MMLAB@CUHK, Shatin, Hong Kong


What is WIDER?

WIDER is a dataset for complex event recognition from static images. As of v0.1, it contains 61 event categories and around 50574 images annotated with event class labels. We provide a split of 50% for training and 50% for testing.

Sample Images



Please use this link for downloading.


This section list the experimental results on WIDER.

Event Classification

Under this setting, the training set of 25275 images will be used to train a event recognition system which classifies an input image into 61 event classes. The test set will be used to evaluate the performance of the system based on its mean recognition accuracy.

Method Mean Accuracy (%) Per-class Results Models
Baseline CNN* 39.67 Link Caffe Models
CNN Deep Channel Fusing 42.42 Link Caffe Models

* the Baseline CNN is finetuned from a well-known AlexNet model pretrained on ImageNet.

You are welcome to submit your results to be listed here.

Citing WIDER

Please cite the following paper in you publication if WIDER helps your research

  title={Recognize Complex Events from Static Images by Fusing Deep Channels},
  author={Xiong, Yuanjun and Zhu, Kai and Lin, Dahua and Tang, Xiaoou},
  booktitle={Computer Vision and Pattern Recognition (CVPR), 2015 IEEE Conference on},


For questions and result submission, please contact Yuanjun Xiong at yjxiong@ie.cuhk.edu.hk