search

KunpengLi1994 / VSRN

PyTorch code for ICCV'19 paper "Visual Semantic Reasoning for Image-Text Matching"
288 stars 47 forks source link

when evaluate, something about models #5

Closed IRIS00123 closed 4 years ago

IRIS00123 commented 4 years ago

def evalrank(model_path, model_path2, data_path=None, split='dev', fold5=False):

there are two models, but when train, save only one model. what are the two models? Thank you very much !

KunpengLi1994 commented 4 years ago

Hi,

Thanks for your question. As stated in Section 4.3 of the paper, we follow recent state-of-the-art method SCAN to report results by combing scores of two trained models. In our case, you can train two times to get two models.