penghu-cs

It is an interesting finding. We have not encountered this phenomenon in our experiments. It is intriguing to explore further. Thanks a lot for your attention.

Sorry for the late reply. I believe that the inconsistency you mentioned will not significantly impact the performance. When calculating the Hamming similarity (inner product), the inconsistency caused by values...

Thank you for reaching out and for your interest in our work. To address your query regarding the reproduction of results from Table 4 for instance-retrieval on Flickr30K, let me...

Dear Mikel, Thank you for reaching out with your query. Implementing instance-level recall evaluation on top of the UCCH code is indeed a feasible approach. While the standalone UCCH implementation...

我没遇到过类似的错误。请问是不是数据集的名不对？

抱歉，我们没有尝试过在cpu上跑。您可以查查相关的文档，看看有没有解决方法。

感谢您指出问题。可以把“[0: k]”删除，或者将“k = numcases”改为“k = dist.shape[1]”，即为map@all。

您好，UCCH/data/README.md 下有全部数据集的下载链接，感谢您的关注。

将src/cmdataset.py文件里，三个路径里的-rand去掉即可。抱歉，上传文件的时候修改了文件名，忘了更新代码。 关于raw data，可以增大warmup_epoch即可，比如：3、5等。

font{ line-height: 1.6; } ul,ol{ padding-left: 20px; list-style-position: inside; } 下载链接如下：MIRFLICKR25K: link: https://pan.baidu.com/s/1o5jSliFjAezBavyBOiJxew password: 8dub祝好，鹏 penghu.ml ***@***.*** ---- 回复的原邮件 ---- 发件人 ***@***.***> 发送日期 2024年01月25日 10:48 收件人 ***@***.***> 抄送人 ***@***.***> 主题...