cd refer && make
- download refcoco+ expressions and annotations from http://tlberg.cs.unc.edu/licheng/referit/data/refcoco+.zip
mkdir data
and extractrefcoco+.zip
intodata
- download and extract
coco_bbox.zip
intorefer
- download refcoco+ image dataset (3.3G - this may take a long time)
python download_dataset.py
- extract vocabularies
python extract_vocabulary.py
- extract data from refcoco+
python parse_data.py
- extract bounding box proposal
python parse_bbox.py
cd ..
and train the model according to ronghang/text_objseg, only change pathexp-referit
intorefer
m-just / text_objseg_coco Goto Github PK
View Code? Open in Web Editor NEWrefcoco+ adaptation of ronghanghu/text_objseg (https://github.com/ronghanghu/text_objseg)
License: Other