/eval_cap

Improved evaluation codes for common visual captioning metrics.

Primary LanguagePython

Visual Caption Evaluation

Improved evaluation codes for common visual captioning metrics (base on coco-caption).

  1. bleu, meteor, rouge, cider, spice
  2. supporting python3
  3. faster cider with cache for reference captions

In order to use spice metric, you should download stanford-corenlp-3.6.0-models.jar into the spice/lib directory.

wget http://nlp.stanford.edu/software/stanford-corenlp-full-2015-12-09.zip
unzip stanford-corenlp-full-2015-12-09
cd stanford-corenlp-full-2015-12-09
mv stanford-corenlp-full-2015-12-09/stanford-corenlp-3.6.0-models.jar spice/lib
mv stanford-corenlp-full-2015-12-09/stanford-corenlp-3.6.0.jar spice/lib