kywen1119
commented
3 years ago 感谢您的关注！ Two-models ensemble是把同一个网络训练两次，分别保存为两个模型（训练时由于随机种子训练结果也不同），分别推理得到image与text的相似度矩阵（在evaluation.py文件中会保存为.npy文件），将两个矩阵求平均得到最终测试的相似度矩阵。 简单地说，基于bert的模型训练两次进行ensemble；基于gru的模型训练两次进行ensemble。这两个是分开的。
Thank you for your attention! For two-models ensemble, we train the same network twice and save them as two models (due to different random seed training results during training), and infers the similarity matrix between image and text (which will be saved in the evaluation.py file as .npy file), average the two matrices to get the final test similarity matrix. Generally, the bert-based model is trained twice for ensemble; the gru-based model is trained twice for ensemble. These two are separate.
xixiareone
你好，我想请教论文表1中的Two-Models Ensemble这一实验，不是很能理解，想请教一下作者，是哪两种model进行ensemble呢，以及对于bert和gru又如何做不同的ensemble？
非常感谢！