SPICE
SPICE copied to clipboard
SPICE on another language?
Hi all!
I want to do SPICE evaluation on Arabic Captions specifically. In the 'spice-1.0.jar' file there is a README file that says:
Dependencies
- Stanford CoreNLP 3.6.0
- Stanford Scene Graph Parser
- Meteor 1.5 (for synset matching)
Since the the version of Stanford CoreNLP is not compatible with Arabic, I wonder if I can:
- Replace Stanford Scene Graph Parser with the latest version, updated to work with CoreNLP 4.2.0
- Replace CoreNLP 3.6.0 with 4.2.0 with Arabic models
I actually tried to download and replace these files in the COCO-caption API, but with no improved SPICE scores...
Do you think this task is possible or no?