We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
limiao: 目前都是固定Image Model的
limiao: 目前Adam正在尝试0.01,0.001和0.0001三种学习率
limiao: 目前SGD正在尝试了两种0.1和0.01两种学习率,decay factor还是0.6。 仅仅从训练集上的cider score变化看,“好像”0.1的学习率好一点,这个结论还有待继续验证