在一些深度哈希里面，都是将cifar10分成trainset-5000，queryset-1000，database-（54000）然后仅仅在trainset-5000上训练。
在您的工作做，是把cifar10分成queryset/database 然后在database上训练，并得到特征提取网络和database的二值化表示，我想问的是假如我在实际应用中换了database，或者database中没有相应的标注。这样应该大大限制了该工作的应用吧

For the ADSH_pytorch, in the ADSH_CIFAR_10.py, I find you random sample the training set from database, the codes is as follows,

for iter in range(max_iter):
        iter_time = time.time()
        '''
        sampling and construct similarity matrix
        '''
        select_index = list(np.random.permutation(range(num_database)))[0: num_samples]
        _sampler = subsetsampler.SubsetSampler(select_index)
        trainloader = DataLoader(dset_database, batch_size=batch_size,
                                 sampler=_sampler,
                                 shuffle=False,
                                 num_workers=4)

However, follow your paper, the training set is randomly sample 500 images for each class. Such that there are 5000 training samples. so in your program there is a list "train_img.txt" with 5000 lines for training.

The problem is I cannot find any codes about loading the train_img.txt ?? can you introduce it clearly?

Hi, it would be nice to have a license in the repo in case we want to publish our work using some of this code. Ideally MIT or Apache 2.0. Thanks!

您好，我是一名大学生，最近在看您的这篇论文，感觉特别棒。另外，看到您用MATLAB实现了DHN，请问您能分享下代码么？万分感谢！原作者的代码是caffe的，不太熟悉。

Hi @jiangqy,
I am trying to test your model with another loss function for the NUSWIDE dataset, the BCEWithLogitsLoss, however I'm having problem with the size of the input and of the labels. It's giving this error:

ValueError: Target size (torch.Size([64, 21])) must be the same as input size (torch.Size([64, 12]))

I do not understand why the input and the labels should be the same size. Can you help me?

Hello! I have an question that I can get the NUS-WIDE.mat from your matlab implement but there is no nus-wide images dataset for pytorch implement. Can the author share the Nus-wide image dataset ?
Thank you very much !

https://github.com/jiangqy/ADSH-AAAI2018/blob/master/ADSH_pytorch/log/log-ADSH-nuswide-17-12-16-09-15-01/log.log，this one .. I am confused about this.

您好，我下载了您提供的百度网盘链接上的NUS-WIDE数据集，但是我分卷解压时一直报错，请问下您是如何解压的？数据集应该是没有问题吧？谢谢了！

Dear Author, Hello, I like your work very much, but I have a question to ask you. How is TOP5K calculated in this paper? Does it mean the average retrieval accuracy of the first 5000 images? Thank you very much and look forward to receiving your reply.

function [dataset, param] = load_data(dataname)
switch dataname
case 'CIFAR-10'
load ./data/CIFAR-10.mat LAll IAll param;
case 'NUS-WIDE'
load ./data/NUS-WIDE.mat LAll IAll param;
case 'MS-COCO'
load ./data/MS-COCO.mat LAll IAll param;
param.indexRetrieval = param.indexDatabase;
end
dataset.IAll = IAll;
dataset.LAll = LAll;
end
想请问这三个数据集的./data/CIFAR-10.mat LAll IAll param是什么呢？

你好，我在复现您的工作时，使用nuswide数据集成功复现出了论文中的效果，但是使用coco数据集时没能得到论文中的效果，且有较大差距。因此，我想请问您在cifar10以及coco数据集上进行实验时的相关参数，或者还有其他需要注意和调整的地方也请您一并告知。
感谢~