flickr8kcn

This page hosts Flickr8K-CN, a bilingual extension of the popular Flickr8K set, used for evaluating image captioning in a cross-lingual setting.

• Chinese sentences written by native Chinese speakers
• Chinese sentences generated by Baidu translation
  • icmr2016 version
  • version 20160815
• Chinese sentences generated by Google translation
  • icmr2016 version
  • version 20160816
• Chinese sentences generated by human translation
• Original English sentences
• Data partition
  • imageids of 6K training images
  • imageids of 1K validation images
  • imageids of 1K test images
• Image features
  • 1,024-dim GoogleNet pool5 layer (read by bigfile.py)

Chinese sentences	Flickr8k-train	Flickr8k-val	Flickr8k-test
machine translation (google)	+	+	+
machine translation (baidu)	+	+	+
human translation	–	–	+
human written	+	+	+

Reference