blue-blue272

Results 15 comments of blue-blue272

@luckycookiecookie I also can not reproduce the performance. Can you solve it?

You can refer to 'https://github.com/kjunelee/MetaOptNet’ to download the tieredImageNet and try again.

Our method is based on the prototype network. In the testing phase, the classifier is not used, we only use the feature extractor part.

The test transductive is for the few-shot classification in the transductive setting. You can find the details in the **Transductive Inference** part of Section 4 in our paper. The meta-learner...

> 您好: 您的代码很简洁,方法的效果也是sota。 但是我想请教一下,您在meta-train之前是否在base class上做过预训练呢?我直接运行您提供的train.py(所有参数均未改动),得到的结果只有百分之二十多?不知道是哪里出了问题,希望能得到您的回复,谢谢! 没有经过预训练。直接运行应该能得出结果的。可能是数据输入格式的问题?

Yes, I understand. Thanks for replying!

I have another question. When you pretrain the model, you set the train loader as : ` train_dataset = loader.ImageLoader( args.data, transforms.Compose([ transforms.RandomResizedCrop(224), transforms.RandomHorizontalFlip(), transforms.ToTensor(), normalize, ]), train=True)` I argue...

> > 你好,我想问一下在第一阶段训练ASR和AAC的过程中的具体的训练设置是什么,使用的epoch base还是step base,大概训练了多少个epochs/steps呢? > > 我们的训练都是 step base 的,数据在 paper 中有提到,32 卡大概训练了 8w 步左右吧 您好,有几个训练的问题想请假一下: 1. 第一阶段是32卡,单卡batch size设置为8, lr按照config的设置为3e-5吗? 2. 第二阶段的训练配置(lr, batchsize等)和第一阶段一样吗?是从第一阶段的模型初始化,用不同的数据进行训练吗?

Thanks. I get that.

@ChenFengYe I have another question, how long about the training time of stage1, stage2 and stage3?