Results in DCMH
Opened this issue · 11 comments
Hi,
Thanks for your brilliant work for reproducing these methods. But when I follow your readme and train a 64bit DCMH on MIRFLICKR25K, the best results is 5 points less than paper(for i->t, the paper is 74.8 and reproducing is 69.4). Below is the screen prints of logs.
Thanks in advance for your reply.
The loss looks normal, maybe caused by the difference in batch-size.
This is my result in 128 batch-size.
The loss looks normal, maybe caused by the difference in batch-size.
This is my result in 128 batch-size.
I guess you are right! When I create the yaml of DCMH, I only change the method of the default yaml where the batch size is remains 64. I think batch size 128 will solve my problem. I do appreciate your help.
Hi,
Thanks for your brilliant work for reproducing these methods. But when I follow your readme and train a 64bit DCMH on MIRFLICKR25K, the best results is 5 points less than paper(for i->t, the paper is 74.8 and reproducing is 69.4). Below is the screen prints of logs.
Thanks in advance for your reply.
Hi,
Thanks for your brilliant work for reproducing these methods. But when I follow your readme and train a 64bit DCMH on MIRFLICKR25K, the best results is 5 points less than paper(for i->t, the paper is 74.8 and reproducing is 69.4). Below is the screen prints of logs.
Thanks in advance for your reply.
hello
Can I ask you a few questions?
Does this code need to download its own data set when it runs?Where should it be placed?I don't see the description in relevant files like README
Thank you very much
Hi,
Thanks for your brilliant work for reproducing these methods. But when I follow your readme and train a 64bit DCMH on MIRFLICKR25K, the best results is 5 points less than paper(for i->t, the paper is 74.8 and reproducing is 69.4). Below is the screen prints of logs.
Thanks in advance for your reply.
Hi,
Thanks for your brilliant work for reproducing these methods. But when I follow your readme and train a 64bit DCMH on MIRFLICKR25K, the best results is 5 points less than paper(for i->t, the paper is 74.8 and reproducing is 69.4). Below is the screen prints of logs.
Thanks in advance for your reply.
hello
Can I ask you a few questions?
Does this code need to download its own data set when it runs?Where should it be placed?I don't see the description in relevant files like README
Thank you very much
You should download the dataset yourself, and write the path where you save dataset on yaml file(see the red mark in the picture). You can create a yaml for a specific method by copying ./scripts/default_config.yaml
Hi,
Thanks for your brilliant work for reproducing these methods. But when I follow your readme and train a 64bit DCMH on MIRFLICKR25K, the best results is 5 points less than paper(for i->t, the paper is 74.8 and reproducing is 69.4). Below is the screen prints of logs.
Thanks in advance for your reply.
Hi,
Thanks for your brilliant work for reproducing these methods. But when I follow your readme and train a 64bit DCMH on MIRFLICKR25K, the best results is 5 points less than paper(for i->t, the paper is 74.8 and reproducing is 69.4). Below is the screen prints of logs.
Thanks in advance for your reply.
hello
Can I ask you a few questions?
Does this code need to download its own data set when it runs?Where should it be placed?I don't see the description in relevant files like README
Thank you very muchYou should download the dataset yourself, and write the path where you save dataset on yaml file(see the red mark in the picture). You can create a yaml for a specific method by copying ./scripts/default_config.yaml
Thank you very much for your reply
Please forgive me for being a beginner in this field
I am also a little confused that I have downloaded the MAT file of MirflckR25K (red box in the picture).
Should I also need to download the mirflckR25K data set image file?Where should I put it?
I beg your criticism and correction
Looking forward to your reply
thank you
I am also a little confused that I have downloaded the MAT file of MirflckR25K (red box in the picture).
Should I also need to download the mirflckR25K data set image file?Where should I put it?
I beg your criticism and correction
Looking forward to your reply
thank you
The .mat files provided by author only include label information. So you should download the mirflckr25K data set yourself. The img_dir path in config.yaml is the place where you put the data set.
我也有点困惑，因为我下载了MirflckR25K的MAT文件（图片中的红色框）。
我还需要下载mirflckR25K数据集图像文件吗？应该放在哪里？
谨请您批评指正。
期待您的答复
谢谢
作者提供的.mat文件仅包含标签信息。因此，您应该自己下载mirflckr25K数据集。config.yaml中的img_dir路径是放置数据集的位置。
Thank you very much indeed
I'll give it a try
I am also a little confused that I have downloaded the MAT file of MirflckR25K (red box in the picture).
Should I also need to download the mirflckR25K data set image file?Where should I put it?
I beg your criticism and correction
Looking forward to your reply
thank youThe .mat files provided by author only include label information. So you should download the mirflckr25K data set yourself. The img_dir path in config.yaml is the place where you put the data set.
Hello, May I ask you again
The situation shown in the picture has been maintained for some time
Does that mean it has run successfully?
If so, how much longer will it take to get results?
Looking forward to your reply
thank you
我也有点困惑，因为我下载了MirflckR25K的MAT文件（图片中的红色框）。
我还需要下载mirflckR25K数据集图像文件吗？应该放在哪里？
谨请您批评指正。
期待您的答复
谢谢
作者提供的.mat文件仅包含标签信息。因此，您应该自己下载mirflckr25K数据集。config.yaml中的img_dir路径是放置数据集的位置。
Thank you very much indeed
I'll give it a try
请问你们运行的时候，四种检索方式有相等吗？
作者提供的.mat文件仅包含标签信息。因此，您应该自己下载mirflckr25K数据集。config.yaml中的img_dir路径是放置数据集的位置。
您好，请问作者给的mat文件里面的文本部分是处理好的向量，还是跟图片部分一样是一个索引，需要自己下载文本数据集呢？
#8 (comment)
@NovaMind-Z
I have encountered exactly the same problem, has your problem been solved?