search

QingyangZhang / QMF

Quality-aware multimodal fusion on ICML 2023
MIT License
70 stars 3 forks source link

复现时发现代码不完整 #2

Closed wenpeiyu123 closed 6 months ago

wenpeiyu123 commented 8 months ago

作者你好，在使用你们的代码时发现models里没有QMF的代码，以及缺少pytorch_pretrained_bert文件，请问这些缺少部分需要如何获取呢，感谢！

QingyangZhang commented 7 months ago

Sorry for the confusion. The only difference between QMF and late fusion lies in the dynamic fusion weights. Therefore, QMF (as implemented in train_df.py and eval_df.py) simply uses late_fusion.py from the 'models' directory. We will update the code to avoid any confusion and will upload the pretrained model as soon as possible.

Xinghetayue commented 7 months ago

Sorry for the confusion. The only difference between QMF and late fusion lies in the dynamic fusion weights. Therefore, QMF (as implemented in train_df.py and eval_df.py) simply uses late_fusion.py from the 'models' directory. We will update the code to avoid any confusion and will upload the pretrained model as soon as possible.

waitting for it