search

kuanghuei / SCAN

PyTorch source code for "Stacked Cross Attention for Image-Text Matching" (ECCV 2018)
Apache License 2.0
549 stars 113 forks source link

No i2t Avg + t2i LSE method #27

Open ahagary opened 4 years ago

ahagary commented 4 years ago

hello, kuanghui I have read you paper and code carefully and I am interested in your approach, I found that there are the results of (i2t Avg + t2i LSE)method in MSCOCO training model , but there are in your code or readme, so I wanna to ask you about how to realize the (i2t Avg + t2i LSE) method to reproduce the best result in your paper, thank you very much

KunpengLi1994 commented 4 years ago

Authors have stated in the paper that they combined the scores obtained from two models.

I have created a pull request to include the implementation of model ensemble. Hope this can help.

ahagary commented 4 years ago

Authors have stated in the paper that they combined the scores obtained from two models.

I have created a pull request to include the implementation of model ensemble. Hope this can help.

Thank you very much!!! I did not consider about model ensemble before seeing your comment, I allways thought that I should add the two scores to calculate loss in a model before I got your help.