li-xirong/flickr8kcn

A bilingual dataset for image captioning

Flickr8K-CN

Flickr8K-CN is a bilingual (English-to-Chinese) extension of the popular Flickr8K set, used for evaluating image captioning in a cross-lingual setting.

Chinese sentences	Flickr8k-train	Flickr8k-val	Flickr8k-test
human written	✅	✅	✅
human translation	❌	❌	✅
machine translation (baidu)	✅	✅	✅
machine translation (google)	✅	✅	✅

Data

Sentences

1. Original English sentences
2. Chinese sentences written by native Chinese speakers
3. Chinese sentences generated by Baidu translation (icmr2016 version, version 20160815)
4. Chinese sentences generated by Google translation (icmr2016 version, version 20160816)
5. Chinese sentences generated by human translation (only the test set is covered)

Dataset split

• imageids of 6K training images, 1k validation images, 1k test images

Image features

1. 1,024-dim GoogleNet pool5, read by bigfile.py

Citations

1. Xirong Li, Weiyu Lan, Jianfeng Dong, Hailong Liu, Adding Chinese Captions to Images, ACM ICMR 2016