search

sibeiyang / sgmn

Graph-Structured Referring Expressions Reasoning in The Wild, In CVPR 2020, Oral.
MIT License
116 stars 15 forks source link

Scripts to generate json files #5

Open thaolmk54 opened 4 years ago

thaolmk54 commented 4 years ago

Hi,

Thank you for the great work. I am wondering if you can share scripts to generate _sgs.json and _sg_seqs.json? I would like to test your method on a customized dataset.

Regards,

sibeiyang commented 4 years ago

Hi @thaolmk54 ,

Thanks for your interest. The scene graphs are generated by Stanford Scene Graph Parser (https://nlp.stanford.edu/software/scenegraph-parser.shtml). And the _sg_seqs.json are generated by reorganizing the scene graphs.

thaolmk54 commented 4 years ago

Hi,

Thanks for the response. I was trying to use the Stanford Scene Graph Parser but the output doesn't look the same as what is in the json file you provided. It is not even close. Have you used a java script to generate the sgs.json? If I read it correctly, there is dependency info ("det", "amod'' etc) associated with each word in a referring expression in sgs.json. I guess you have done some post-processing before writing the results to the json file. Very much appreciated if you can share the scripts you used. Plus it's almost impossible to replicate exactly how did you reorganize the scene graphs by reading your paper alone.