About COMPOSITE dataset in the paper

peteanderson80 / SPICE

Semantic Propositional Image Caption Evaluation

http://panderson.me/spice

GNU Affero General Public License v3.0

132 stars 31 forks source link

About COMPOSITE dataset in the paper #10

Open Yudezhi opened 3 years ago

Yudezhi commented 3 years ago

Hi: I am confused about the COMPOSITE dataset proposed in the paper. How 19985 annotated sentences choose from the source file download from [here] (https://www.dropbox.com/s/z1l4o3j0gj8yyp8/AMT_eval.zip?dl=0). The number of sentences for an image is 4 in 30k and MScoco instead of 3. Thanks~