Is there overfitting?

Question

Is there overfitting?

Closed this issue 3 years ago · 20 comments

Hello, I train the Track2 Single-modal with my dataset which include 5000+ real and 5000+ fake pics. The epoch is 60, and train log as follow:

Oulu-NPU, P1:
train from scratch!
epoch:1, Train: Absolute_Depth_loss= 0.2492, Contrastive_Depth_loss= 0.0112
epoch:2, Train: Absolute_Depth_loss= 0.1995, Contrastive_Depth_loss= 0.0083
epoch:3, Train: Absolute_Depth_loss= 0.1734, Contrastive_Depth_loss= 0.0087
epoch:4, Train: Absolute_Depth_loss= 0.1561, Contrastive_Depth_loss= 0.0088
epoch:5, Train: Absolute_Depth_loss= 0.1435, Contrastive_Depth_loss= 0.0089
epoch:6, Train: Absolute_Depth_loss= 0.1334, Contrastive_Depth_loss= 0.0089
epoch:7, Train: Absolute_Depth_loss= 0.1266, Contrastive_Depth_loss= 0.0090
epoch:8, Train: Absolute_Depth_loss= 0.1210, Contrastive_Depth_loss= 0.0091
epoch:9, Train: Absolute_Depth_loss= 0.1133, Contrastive_Depth_loss= 0.0090
epoch:10, Train: Absolute_Depth_loss= 0.1085, Contrastive_Depth_loss= 0.0091
epoch:11, Train: Absolute_Depth_loss= 0.1039, Contrastive_Depth_loss= 0.0092
epoch:12, Train: Absolute_Depth_loss= 0.0988, Contrastive_Depth_loss= 0.0092
epoch:13, Train: Absolute_Depth_loss= 0.0939, Contrastive_Depth_loss= 0.0092
epoch:14, Train: Absolute_Depth_loss= 0.0907, Contrastive_Depth_loss= 0.0091
epoch:15, Train: Absolute_Depth_loss= 0.0875, Contrastive_Depth_loss= 0.0092
epoch:16, Train: Absolute_Depth_loss= 0.0833, Contrastive_Depth_loss= 0.0091
epoch:17, Train: Absolute_Depth_loss= 0.0810, Contrastive_Depth_loss= 0.0091
epoch:18, Train: Absolute_Depth_loss= 0.0791, Contrastive_Depth_loss= 0.0090
epoch:19, Train: Absolute_Depth_loss= 0.0768, Contrastive_Depth_loss= 0.0090
epoch:20, Train: Absolute_Depth_loss= 0.0622, Contrastive_Depth_loss= 0.0084
epoch:21, Train: Absolute_Depth_loss= 0.0591, Contrastive_Depth_loss= 0.0085
epoch:22, Train: Absolute_Depth_loss= 0.0562, Contrastive_Depth_loss= 0.0084
epoch:23, Train: Absolute_Depth_loss= 0.0543, Contrastive_Depth_loss= 0.0085
epoch:24, Train: Absolute_Depth_loss= 0.0532, Contrastive_Depth_loss= 0.0084
epoch:25, Train: Absolute_Depth_loss= 0.0511, Contrastive_Depth_loss= 0.0083
epoch:26, Train: Absolute_Depth_loss= 0.0497, Contrastive_Depth_loss= 0.0083
epoch:27, Train: Absolute_Depth_loss= 0.0486, Contrastive_Depth_loss= 0.0083
epoch:28, Train: Absolute_Depth_loss= 0.0465, Contrastive_Depth_loss= 0.0082
epoch:29, Train: Absolute_Depth_loss= 0.0455, Contrastive_Depth_loss= 0.0082
epoch:30, Train: Absolute_Depth_loss= 0.0445, Contrastive_Depth_loss= 0.0081
epoch:31, Train: Absolute_Depth_loss= 0.0441, Contrastive_Depth_loss= 0.0082
epoch:32, Train: Absolute_Depth_loss= 0.0424, Contrastive_Depth_loss= 0.0081
epoch:33, Train: Absolute_Depth_loss= 0.0413, Contrastive_Depth_loss= 0.0080
epoch:34, Train: Absolute_Depth_loss= 0.0416, Contrastive_Depth_loss= 0.0080
epoch:35, Train: Absolute_Depth_loss= 0.0401, Contrastive_Depth_loss= 0.0079
epoch:36, Train: Absolute_Depth_loss= 0.0396, Contrastive_Depth_loss= 0.0080
epoch:37, Train: Absolute_Depth_loss= 0.0391, Contrastive_Depth_loss= 0.0079
epoch:38, Train: Absolute_Depth_loss= 0.0389, Contrastive_Depth_loss= 0.0079
epoch:39, Train: Absolute_Depth_loss= 0.0355, Contrastive_Depth_loss= 0.0077
epoch:40, Train: Absolute_Depth_loss= 0.0307, Contrastive_Depth_loss= 0.0073
epoch:41, Train: Absolute_Depth_loss= 0.0298, Contrastive_Depth_loss= 0.0072
epoch:42, Train: Absolute_Depth_loss= 0.0289, Contrastive_Depth_loss= 0.0072
epoch:43, Train: Absolute_Depth_loss= 0.0286, Contrastive_Depth_loss= 0.0071
epoch:44, Train: Absolute_Depth_loss= 0.0278, Contrastive_Depth_loss= 0.0071
epoch:45, Train: Absolute_Depth_loss= 0.0266, Contrastive_Depth_loss= 0.0070
epoch:46, Train: Absolute_Depth_loss= 0.0269, Contrastive_Depth_loss= 0.0070
epoch:47, Train: Absolute_Depth_loss= 0.0264, Contrastive_Depth_loss= 0.0070
epoch:48, Train: Absolute_Depth_loss= 0.0266, Contrastive_Depth_loss= 0.0070
epoch:49, Train: Absolute_Depth_loss= 0.0250, Contrastive_Depth_loss= 0.0069
epoch:50, Train: Absolute_Depth_loss= 0.0258, Contrastive_Depth_loss= 0.0070
epoch:51, Train: Absolute_Depth_loss= 0.0245, Contrastive_Depth_loss= 0.0068
epoch:52, Train: Absolute_Depth_loss= 0.0237, Contrastive_Depth_loss= 0.0067
epoch:53, Train: Absolute_Depth_loss= 0.0241, Contrastive_Depth_loss= 0.0069
epoch:54, Train: Absolute_Depth_loss= 0.0233, Contrastive_Depth_loss= 0.0068
epoch:55, Train: Absolute_Depth_loss= 0.0236, Contrastive_Depth_loss= 0.0067
epoch:56, Train: Absolute_Depth_loss= 0.0228, Contrastive_Depth_loss= 0.0067
epoch:57, Train: Absolute_Depth_loss= 0.0226, Contrastive_Depth_loss= 0.0067
epoch:58, Train: Absolute_Depth_loss= 0.0218, Contrastive_Depth_loss= 0.0066
epoch:59, Train: Absolute_Depth_loss= 0.0216, Contrastive_Depth_loss= 0.0066
epoch:60, Train: Absolute_Depth_loss= 0.0193, Contrastive_Depth_loss= 0.0063

when I use the saved model file to test. I find some images perform poorly, Fake img is recongize to real. The accuracy about 24%. Is there overfitting? or another reason?

Another question, the pics which use to train is raw image or crop face? Is there any difference between them?

Thank you very much.

Answer 1 · 2020-06-16T13:25:14.000Z

I train the whole model based on Keras API, got no overfitting problem, just found that the model author proposed in their paper is slow. Anyway, maybe you should do more data augmentation, like gamma correction, saturation correction.

Answer 2 · 2020-06-17T02:53:43.000Z

@xuhangxuhang Thank you.

Could you tell me which dataset you use to train? After training, how does your model performance at the data out of the dataset you train?

Answer 3 · 2020-06-21T07:06:36.000Z

@xuhangxuhang Thank you.

Could you tell me which dataset you use to train? After training, how does your model performance at the data out of the dataset you train?

I got good results inside Oulu, CASIA-FASD, and Replayattack, but cross-test results of CASIA-FASD and Replayattack is bad.

Answer 4 · 2020-06-22T00:53:29.000Z

@xuhangxuhang Thank you.
Could you tell me which dataset you use to train? After training, how does your model performance at the data out of the dataset you train?

I got good results inside Oulu, CASIA-FASD, and Replayattack, but cross-test results of CASIA-FASD and Replayattack is bad.

Thank you very much.

Answer 5 · 2020-07-15T01:41:42.000Z

@xuhangxuhang Thank you.

Could you tell me which dataset you use to train? After training, how does your model performance at the data out of the dataset you train?

I got good results inside Oulu, CASIA-FASD, and Replayattack, but cross-test results of CASIA-FASD and Replayattack is bad.

can you please tell me, that how did you use the Oulu dataset, because in the dataset I found the videos.
but in the source code they required the map_dir, even the train, test, dev required the images.

Answer 6 · 2020-07-15T08:27:23.000Z

@ysm022

Can you please tell me about the map_dir/
It will be a great help

Thanks in advance

Answer 7 · 2020-07-26T03:18:10.000Z

Hi，sorry I replay this email so late, sincerely apologize for that. In Oulu-NPU dataset, the code provider sent single image into their network. So what you should do is, write a generator(in pytorch write a class as the code in the Github Repo, if you use TF or Keras, a simple function is enough), the generator should yeild image(face iamge) and the corresponding depth map. In training stage, get average output score of each sample, in evaluation stage, get prediction scores of samples from original video and average them as the final prediction result of single video. (ps: I am happy to reply your question, and my English is not that good, if I confuse you please ask again, I will try my best to answer.) Best wish.

…

-----原始邮件----- 发件人:punitha-valli <notifications@github.com> 发送时间:2020-07-15 09:42:03 (星期三) 收件人: ZitongYu/CDCN <CDCN@noreply.github.com> 抄送: xuhangxuhang <hangxu@my.swjtu.edu.cn>, Mention <mention@noreply.github.com> 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) @xuhangxuhang Thank you. Could you tell me which dataset you use to train? After training, how does your model performance at the data out of the dataset you train? I got good results inside Oulu, CASIA-FASD, and Replayattack, but cross-test results of CASIA-FASD and Replayattack is bad. can you please tell me, that how did you use the Oulu dataset, because in the dataset I found the videos. but in the source code they required the map_dir, even the train, test, dev required the images. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

Answer 8 · 2020-07-27T05:08:03.000Z

Thank you for your response I have Oulu dataset, but it's fully video. Then I don't understand about the map_dir. Then the image path , because I have videos in Oulu dataset,.... What should I do for the train_image, test_image, dev_image, map_dir, Can you please help me? Thank you so much .

…

On Sun 26 Jul, 2020, 11:18 AM xuhangxuhang, ***@***.***> wrote: Hi，sorry I replay this email so late, sincerely apologize for that. In Oulu-NPU dataset, the code provider sent single image into their network. So what you should do is, write a generator(in pytorch write a class as the code in the Github Repo, if you use TF or Keras, a simple function is enough), the generator should yeild image(face iamge) and the corresponding depth map. In training stage, get average output score of each sample, in evaluation stage, get prediction scores of samples from original video and average them as the final prediction result of single video. (ps: I am happy to reply your question, and my English is not that good, if I confuse you please ask again, I will try my best to answer.) Best wish. -----原始邮件----- 发件人:punitha-valli ***@***.***> 发送时间:2020-07-15 09:42:03 (星期三) 收件人: ZitongYu/CDCN ***@***.***> 抄送: xuhangxuhang ***@***.***>, Mention < ***@***.***> 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) @xuhangxuhang Thank you. Could you tell me which dataset you use to train? After training, how does your model performance at the data out of the dataset you train? I got good results inside Oulu, CASIA-FASD, and Replayattack, but cross-test results of CASIA-FASD and Replayattack is bad. can you please tell me, that how did you use the Oulu dataset, because in the dataset I found the videos. but in the source code they required the map_dir, even the train, test, dev required the images. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#21 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA> .

Answer 9 · 2020-07-27T06:03:15.000Z

What you should do is split the title video into single frames, for every live frame yield depth map, for each spoof frame just yield an all zero matrix, mapdir is the label save folder. Is this clear? hangxu 邮箱：hangxu@my.swjtu.edu.cn 签名由网易邮箱大师定制 On 07/27/2020 13:08, punitha-valli wrote: Thank you for your response I have Oulu dataset, but it's fully video. Then I don't understand about the map_dir. Then the image path , because I have videos in Oulu dataset,.... What should I do for the train_image, test_image, dev_image, map_dir, Can you please help me? Thank you so much . On Sun 26 Jul, 2020, 11:18 AM xuhangxuhang, <notifications@github.com> wrote: > Hi，sorry I replay this email so late, sincerely apologize for that. > > > > > In Oulu-NPU dataset, the code provider sent single image into their > network. So what you should do is, write a generator(in pytorch write a > class as the code in the Github Repo, if you use TF or Keras, a simple > function is enough), the generator should yeild image(face iamge) and the > corresponding depth map. In training stage, get average output score of > each sample, in evaluation stage, get prediction scores of samples from > original video and average them as the final prediction result of single > video. > > > > > (ps: I am happy to reply your question, and my English is not that good, > if I confuse you please ask again, I will try my best to answer.) > > Best wish. > > > > > -----原始邮件----- > 发件人:punitha-valli <notifications@github.com> > 发送时间:2020-07-15 09:42:03 (星期三) > 收件人: ZitongYu/CDCN <CDCN@noreply.github.com> > 抄送: xuhangxuhang <hangxu@my.swjtu.edu.cn>, Mention < > mention@noreply.github.com> > 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could you tell me which dataset you use to train? After training, how does > your model performance at the data out of the dataset you train? > > I got good results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test results of CASIA-FASD and Replayattack is bad. > > can you please tell me, that how did you use the Oulu dataset, because in > the dataset I found the videos. > but in the source code they required the map_dir, even the train, test, > dev required the images. > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub, or unsubscribe. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#21 (comment)>, or > unsubscribe > <https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA> > . > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

Answer 10 · 2020-07-27T06:34:26.000Z

Can you please share the code for making map_dir ?

…

On Mon 27 Jul, 2020, 2:03 PM xuhangxuhang, ***@***.***> wrote: What you should do is split the title video into single frames, for every live frame yield depth map, for each spoof frame just yield an all zero matrix, mapdir is the label save folder. Is this clear? hangxu 邮箱： ***@***.*** 签名由网易邮箱大师定制 On 07/27/2020 13:08, punitha-valli wrote: Thank you for your response I have Oulu dataset, but it's fully video. Then I don't understand about the map_dir. Then the image path , because I have videos in Oulu dataset,.... What should I do for the train_image, test_image, dev_image, map_dir, Can you please help me? Thank you so much . On Sun 26 Jul, 2020, 11:18 AM xuhangxuhang, < ***@***.***> wrote: > Hi，sorry I replay this email so late, sincerely apologize for that. > > > > > In Oulu-NPU dataset, the code provider sent single image into their > network. So what you should do is, write a generator(in pytorch write a > class as the code in the Github Repo, if you use TF or Keras, a simple > function is enough), the generator should yeild image(face iamge) and the > corresponding depth map. In training stage, get average output score of > each sample, in evaluation stage, get prediction scores of samples from > original video and average them as the final prediction result of single > video. > > > > > (ps: I am happy to reply your question, and my English is not that good, > if I confuse you please ask again, I will try my best to answer.) > > Best wish. > > > > > -----原始邮件----- > 发件人:punitha-valli ***@***.***> > 发送时间:2020-07-15 09:42:03 (星期三) > 收件人: ZitongYu/CDCN < ***@***.***> > 抄送: xuhangxuhang ***@***.***>, Mention < > ***@***.***> > 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could you tell me which dataset you use to train? After training, how does > your model performance at the data out of the dataset you train? > > I got good results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test results of CASIA-FASD and Replayattack is bad. > > can you please tell me, that how did you use the Oulu dataset, because in > the dataset I found the videos. > but in the source code they required the map_dir, even the train, test, > dev required the images. > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub, or unsubscribe. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > < #21 (comment)>, or > unsubscribe > < https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA> > . > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#21 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMM7B2YW6ARN25YF3UIY4NDR5UKDBANCNFSM4N7N7IKA> .

Answer 11 · 2020-07-27T07:06:46.000Z

you can search PRNet iin GitHub, all map labels I have are generated from that code. best wish hangxu 邮箱：hangxu@my.swjtu.edu.cn 签名由网易邮箱大师定制 On 07/27/2020 14:34, punitha-valli wrote: Can you please share the code for making map_dir ? On Mon 27 Jul, 2020, 2:03 PM xuhangxuhang, <notifications@github.com> wrote: > What you should do is split the title video into single frames, for every > live frame yield depth map, for each spoof frame just yield an all zero > matrix, mapdir is the label save folder. Is this clear? hangxu 邮箱： > hangxu@my.swjtu.edu.cn 签名由网易邮箱大师定制 On 07/27/2020 13:08, punitha-valli > wrote: Thank you for your response I have Oulu dataset, but it's fully > video. Then I don't understand about the map_dir. Then the image path , > because I have videos in Oulu dataset,.... What should I do for the > train_image, test_image, dev_image, map_dir, Can you please help me? Thank > you so much . On Sun 26 Jul, 2020, 11:18 AM xuhangxuhang, < > notifications@github.com> wrote: > Hi，sorry I replay this email so late, > sincerely apologize for that. > > > > > In Oulu-NPU dataset, the code > provider sent single image into their > network. So what you should do is, > write a generator(in pytorch write a > class as the code in the Github > Repo, if you use TF or Keras, a simple > function is enough), the generator > should yeild image(face iamge) and the > corresponding depth map. In > training stage, get average output score of > each sample, in evaluation > stage, get prediction scores of samples from > original video and average > them as the final prediction result of single > video. > > > > > (ps: I am > happy to reply your question, and my English is not that good, > if I > confuse you please ask again, I will try my best to answer.) > > Best wish. > > > > > > -----原始邮件----- > 发件人:punitha-valli <notifications@github.com> > > 发送时间:2020-07-15 09:42:03 (星期三) > 收件人: ZitongYu/CDCN < > CDCN@noreply.github.com> > 抄送: xuhangxuhang <hangxu@my.swjtu.edu.cn>, > Mention < > mention@noreply.github.com> > 主题: Re: [ZitongYu/CDCN] Is > there overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could > you tell me which dataset you use to train? After training, how does > your > model performance at the data out of the dataset you train? > > I got good > results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test results > of CASIA-FASD and Replayattack is bad. > > can you please tell me, that how > did you use the Oulu dataset, because in > the dataset I found the videos. > > but in the source code they required the map_dir, even the train, test, > > dev required the images. > > — > You are receiving this because you were > mentioned. > Reply to this email directly, view it on GitHub, or > unsubscribe. > > — > You are receiving this because you commented. > Reply > to this email directly, view it on GitHub > < > #21 (comment)>, or > > unsubscribe > < > https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA> > > . > — You are receiving this because you were mentioned. Reply to this > email directly, view it on GitHub, or unsubscribe. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#21 (comment)>, or > unsubscribe > <https://github.com/notifications/unsubscribe-auth/AMM7B2YW6ARN25YF3UIY4NDR5UKDBANCNFSM4N7N7IKA> > . > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

Answer 12 · 2020-07-27T07:42:13.000Z

Thank you for so much,. I have tried that one , but my dlib in not working properly. The wheel of dlib is not building correctly.

…

On Mon 27 Jul, 2020, 3:07 PM xuhangxuhang, ***@***.***> wrote: you can search PRNet iin GitHub, all map labels I have are generated from that code. best wish hangxu ***@***.*** 签名由网易邮箱大师定制 On 07/27/2020 14:34, punitha-valli wrote: Can you please share the code for making map_dir ? On Mon 27 Jul, 2020, 2:03 PM xuhangxuhang, < ***@***.***> wrote: > What you should do is split the title video into single frames, for every > live frame yield depth map, for each spoof frame just yield an all zero > matrix, mapdir is the label save folder. Is this clear? hangxu 邮箱： > ***@***.*** 签名由网易邮箱大师定制 On 07/27/2020 13:08, punitha-valli > wrote: Thank you for your response I have Oulu dataset, but it's fully > video. Then I don't understand about the map_dir. Then the image path , > because I have videos in Oulu dataset,.... What should I do for the > train_image, test_image, dev_image, map_dir, Can you please help me? Thank > you so much . On Sun 26 Jul, 2020, 11:18 AM xuhangxuhang, < > ***@***.***> wrote: > Hi，sorry I replay this email so late, > sincerely apologize for that. > > > > > In Oulu-NPU dataset, the code > provider sent single image into their > network. So what you should do is, > write a generator(in pytorch write a > class as the code in the Github > Repo, if you use TF or Keras, a simple > function is enough), the generator > should yeild image(face iamge) and the > corresponding depth map. In > training stage, get average output score of > each sample, in evaluation > stage, get prediction scores of samples from > original video and average > them as the final prediction result of single > video. > > > > > (ps: I am > happy to reply your question, and my English is not that good, > if I > confuse you please ask again, I will try my best to answer.) > > Best wish. > > > > > > -----原始邮件----- > 发件人:punitha-valli ***@***.***> > > 发送时间:2020-07-15 09:42:03 (星期三) > 收件人: ZitongYu/CDCN < > ***@***.***> > 抄送: xuhangxuhang ***@***.***>, > Mention < > ***@***.***> > 主题: Re: [ZitongYu/CDCN] Is > there overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could > you tell me which dataset you use to train? After training, how does > your > model performance at the data out of the dataset you train? > > I got good > results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test results > of CASIA-FASD and Replayattack is bad. > > can you please tell me, that how > did you use the Oulu dataset, because in > the dataset I found the videos. > > but in the source code they required the map_dir, even the train, test, > > dev required the images. > > — > You are receiving this because you were > mentioned. > Reply to this email directly, view it on GitHub, or > unsubscribe. > > — > You are receiving this because you commented. > Reply > to this email directly, view it on GitHub > < > #21 (comment)>, or > > unsubscribe > < > https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA> > > . > — You are receiving this because you were mentioned. Reply to this > email directly, view it on GitHub, or unsubscribe. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > < #21 (comment)>, or > unsubscribe > < https://github.com/notifications/unsubscribe-auth/AMM7B2YW6ARN25YF3UIY4NDR5UKDBANCNFSM4N7N7IKA> > . > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#21 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMM7B22TINYGOLLCP62E6A3R5URRJANCNFSM4N7N7IKA> .

Answer 13 · 2020-07-27T08:24:49.000Z

Thank you for your kind response How many frames did you used for one video ?, While converting video into images. How did you saved the image names ?, I have some problem with this . On Mon 27 Jul, 2020, 3:42 PM Punitha Valli, <punithavalli.krv@gmail.com> wrote:

…

Thank you for so much,. I have tried that one , but my dlib in not working properly. The wheel of dlib is not building correctly. On Mon 27 Jul, 2020, 3:07 PM xuhangxuhang, ***@***.***> wrote: > you can search PRNet iin GitHub, all map labels I have are generated > from that code. best wish hangxu ***@***.*** 签名由网易邮箱大师 > 定制 On 07/27/2020 14:34, punitha-valli wrote: Can you please share the code > for making map_dir ? On Mon 27 Jul, 2020, 2:03 PM xuhangxuhang, < > ***@***.***> wrote: > What you should do is split the title > video into single frames, for every > live frame yield depth map, for each > spoof frame just yield an all zero > matrix, mapdir is the label save > folder. Is this clear? hangxu 邮箱： > ***@***.*** 签名由网易邮箱大师定制 > On 07/27/2020 13:08, punitha-valli > wrote: Thank you for your response I > have Oulu dataset, but it's fully > video. Then I don't understand about > the map_dir. Then the image path , > because I have videos in Oulu > dataset,.... What should I do for the > train_image, test_image, dev_image, > map_dir, Can you please help me? Thank > you so much . On Sun 26 Jul, 2020, > 11:18 AM xuhangxuhang, < > ***@***.***> wrote: > Hi，sorry I > replay this email so late, > sincerely apologize for that. > > > > > In > Oulu-NPU dataset, the code > provider sent single image into their > > network. So what you should do is, > write a generator(in pytorch write a > > class as the code in the Github > Repo, if you use TF or Keras, a simple > > function is enough), the generator > should yeild image(face iamge) and the > > corresponding depth map. In > training stage, get average output score of > > each sample, in evaluation > stage, get prediction scores of samples from > > original video and average > them as the final prediction result of > single > video. > > > > > (ps: I am > happy to reply your question, and my > English is not that good, > if I > confuse you please ask again, I will try > my best to answer.) > > Best wish. > > > > > > -----原始邮件----- > > 发件人:punitha-valli ***@***.***> > > 发送时间:2020-07-15 > 09:42:03 (星期三) > 收件人: ZitongYu/CDCN < > ***@***.***> > 抄送: > xuhangxuhang ***@***.***>, > Mention < > > ***@***.***> > 主题: Re: [ZitongYu/CDCN] Is > there > overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could > you > tell me which dataset you use to train? After training, how does > your > > model performance at the data out of the dataset you train? > > I got good > > results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test > results > of CASIA-FASD and Replayattack is bad. > > can you please tell > me, that how > did you use the Oulu dataset, because in > the dataset I > found the videos. > > but in the source code they required the map_dir, > even the train, test, > > dev required the images. > > — > You are > receiving this because you were > mentioned. > Reply to this email > directly, view it on GitHub, or > unsubscribe. > > — > You are receiving > this because you commented. > Reply > to this email directly, view it on > GitHub > < > > #21 (comment)>, or > > > unsubscribe > < > > https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA> > > > . > — You are receiving this because you were mentioned. Reply to this > > email directly, view it on GitHub, or unsubscribe. > > — > You are > receiving this because you commented. > Reply to this email directly, view > it on GitHub > < > #21 (comment)>, or > > unsubscribe > < > https://github.com/notifications/unsubscribe-auth/AMM7B2YW6ARN25YF3UIY4NDR5UKDBANCNFSM4N7N7IKA> > > . > — You are receiving this because you were mentioned. Reply to this > email directly, view it on GitHub, or unsubscribe. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#21 (comment)>, or > unsubscribe > <https://github.com/notifications/unsubscribe-auth/AMM7B22TINYGOLLCP62E6A3R5URRJANCNFSM4N7N7IKA> > . >

Answer 14 · 2020-07-28T11:23:52.000Z

In Oulu-NPU dataset, I randoimly select 20 frames of each video of dev/test subset, for training I balance data by video name first, for example in Oulu-NPU Protocol-1 there are 240 real face video and 960 spoofed ones if my memory is correct, I copy real samples for 4 times and mixed the expaneded real video list and spoofed video list, shuffled them. What I received is a balanced training set. Next randomly select one frame from every balanced video, training them then. I set 10000 samples for an epoch, and batchsize is 20. So For each epoch my model runs 500 steps. That's it. For your 2nd question, what you should do is just save the balanced training name list as a .txt file, the same as Oulu-NPU dataset provided. Sorry I can not show you my code now, I am already graduated, so my code are deleted. Any question, just ask me, I will replay you when I got time. Best wish.

-----原始邮件----- 发件人:punitha-valli <notifications@github.com> 发送时间:2020-07-27 16:25:04 (星期一) 收件人: ZitongYu/CDCN <CDCN@noreply.github.com> 抄送: xuhangxuhang <hangxu@my.swjtu.edu.cn>, Mention <mention@noreply.github.com> 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) Thank you for your kind response How many frames did you used for one video ?, While converting video into images. How did you saved the image names ?, I have some problem with this . On Mon 27 Jul, 2020, 3:42 PM Punitha Valli, <punithavalli.krv@gmail.com> wrote:

Thank you for so much,. I have tried that one , but my dlib in not working properly. The wheel of dlib is not building correctly. On Mon 27 Jul, 2020, 3:07 PM xuhangxuhang, ***@***.***> wrote: > you can search PRNet iin GitHub, all map labels I have are generated > from that code. best wish hangxu ***@***.*** 签名由网易邮箱大师 > 定制 On 07/27/2020 14:34, punitha-valli wrote: Can you please share the code > for making map_dir ? On Mon 27 Jul, 2020, 2:03 PM xuhangxuhang, < > ***@***.***> wrote: > What you should do is split the title > video into single frames, for every > live frame yield depth map, for each > spoof frame just yield an all zero > matrix, mapdir is the label save > folder. Is this clear? hangxu 邮箱： > ***@***.*** 签名由网易邮箱大师定制 > On 07/27/2020 13:08, punitha-valli > wrote: Thank you for your response I > have Oulu dataset, but it's fully > video. Then I don't understand about > the map_dir. Then the image path , > because I have videos in Oulu > dataset,.... What should I do for the > train_image, test_image, dev_image, > map_dir, Can you please help me? Thank > you so much . On Sun 26 Jul, 2020, > 11:18 AM xuhangxuhang, < > ***@***.***> wrote: > Hi，sorry I > replay this email so late, > sincerely apologize for that. > > > > > In > Oulu-NPU dataset, the code > provider sent single image into their > > network. So what you should do is, > write a generator(in pytorch write a > > class as the code in the Github > Repo, if you use TF or Keras, a simple > > function is enough), the generator > should yeild image(face iamge) and the > > corresponding depth map. In > training stage, get average output score of > > each sample, in evaluation > stage, get prediction scores of samples from > > original video and average > them as the final prediction result of > single > video. > > > > > (ps: I am > happy to reply your question, and my > English is not that good, > if I > confuse you please ask again, I will try > my best to answer.) > > Best wish. > > > > > > -----原始邮件----- > > 发件人:punitha-valli ***@***.***> > > 发送时间:2020-07-15 > 09:42:03 (星期三) > 收件人: ZitongYu/CDCN < > ***@***.***> > 抄送: > xuhangxuhang ***@***.***>, > Mention < > > ***@***.***> > 主题: Re: [ZitongYu/CDCN] Is > there > overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could > you > tell me which dataset you use to train? After training, how does > your > > model performance at the data out of the dataset you train? > > I got good > > results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test > results > of CASIA-FASD and Replayattack is bad. > > can you please tell > me, that how > did you use the Oulu dataset, because in > the dataset I > found the videos. > > but in the source code they required the map_dir, > even the train, test, > > dev required the images. > > — > You are > receiving this because you were > mentioned. > Reply to this email > directly, view it on GitHub, or > unsubscribe. > > — > You are receiving > this because you commented. > Reply > to this email directly, view it on > GitHub > < > > #21 (comment)>, or > > > unsubscribe > < > > https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA> > > > . > — You are receiving this because you were mentioned. Reply to this > > email directly, view it on GitHub, or unsubscribe. > > — > You are > receiving this because you commented. > Reply to this email directly, view > it on GitHub > < > #21 (comment)>, or > > unsubscribe > < > https://github.com/notifications/unsubscribe-auth/AMM7B2YW6ARN25YF3UIY4NDR5UKDBANCNFSM4N7N7IKA> > > . > — You are receiving this because you were mentioned. Reply to this > email directly, view it on GitHub, or unsubscribe. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#21 (comment)>, or > unsubscribe > <https://github.com/notifications/unsubscribe-auth/AMM7B22TINYGOLLCP62E6A3R5URRJANCNFSM4N7N7IKA> > . >

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

Answer 15 · 2020-07-29T02:54:38.000Z

Sry I have no line app, I am from China, so you can connect me with Email or WeChat.

-----原始邮件----- 发件人:punitha-valli <notifications@github.com> 发送时间:2020-07-29 10:35:30 (星期三) 收件人: ZitongYu/CDCN <CDCN@noreply.github.com> 抄送: xuhangxuhang <hangxu@my.swjtu.edu.cn>, Mention <mention@noreply.github.com> 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) Thank you so much, Can i ask you a personal question? Can you share your line id, for future reference? My line id: spv.krv Thanks in advance

On Tue, 28 Jul 2020 at 19:24, xuhangxuhang ***@***.***> wrote: In Oulu-NPU dataset, I randoimly select 20 frames of each video of dev/test subset, for training I balance data by video name first, for example in Oulu-NPU Protocol-1 there are 240 real face video and 960 spoofed ones if my memory is correct, I copy real samples for 4 times and mixed the expaneded real video list and spoofed video list, shuffled them. What I received is a balanced training set. Next randomly select one frame from every balanced video, training them then. I set 10000 samples for an epoch, and batchsize is 20. So For each epoch my model runs 500 steps. That's it. For your 2nd question, what you should do is just save the balanced training name list as a .txt file, the same as Oulu-NPU dataset provided. Sorry I can not show you my code now, I am already graduated, so my code are deleted. Any question, just ask me, I will replay you when I got time. Best wish. -----原始邮件----- 发件人:punitha-valli ***@***.***> 发送时间:2020-07-27 16:25:04 (星期一) 收件人: ZitongYu/CDCN ***@***.***> 抄送: xuhangxuhang ***@***.***>, Mention < ***@***.***> 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) Thank you for your kind response How many frames did you used for one video ?, While converting video into images. How did you saved the image names ?, I have some problem with this . On Mon 27 Jul, 2020, 3:42 PM Punitha Valli, ***@***.***> wrote: > Thank you for so much,. > > I have tried that one , but my dlib in not working properly. The wheel of > dlib is not building correctly. > > > On Mon 27 Jul, 2020, 3:07 PM xuhangxuhang, ***@***.***> > wrote: > >> you can search PRNet iin GitHub, all map labels I have are generated >> from that code. best wish hangxu ***@***.*** 签名由网易邮箱大师 >> 定制 On 07/27/2020 14:34, punitha-valli wrote: Can you please share the code >> for making map_dir ? On Mon 27 Jul, 2020, 2:03 PM xuhangxuhang, < >> ***@***.***> wrote: > What you should do is split the title >> video into single frames, for every > live frame yield depth map, for each >> spoof frame just yield an all zero > matrix, mapdir is the label save >> folder. Is this clear? hangxu 邮箱： > ***@***.*** 签名由网易邮箱大师定制 >> On 07/27/2020 13:08, punitha-valli > wrote: Thank you for your response I >> have Oulu dataset, but it's fully > video. Then I don't understand about >> the map_dir. Then the image path , > because I have videos in Oulu >> dataset,.... What should I do for the > train_image, test_image, dev_image, >> map_dir, Can you please help me? Thank > you so much . On Sun 26 Jul, 2020, >> 11:18 AM xuhangxuhang, < > ***@***.***> wrote: > Hi，sorry I >> replay this email so late, > sincerely apologize for that. > > > > > In >> Oulu-NPU dataset, the code > provider sent single image into their > >> network. So what you should do is, > write a generator(in pytorch write a > >> class as the code in the Github > Repo, if you use TF or Keras, a simple > >> function is enough), the generator > should yeild image(face iamge) and the >> > corresponding depth map. In > training stage, get average output score of >> > each sample, in evaluation > stage, get prediction scores of samples from >> > original video and average > them as the final prediction result of >> single > video. > > > > > (ps: I am > happy to reply your question, and my >> English is not that good, > if I > confuse you please ask again, I will try >> my best to answer.) > > Best wish. > > > > > > -----原始邮件----- > >> 发件人:punitha-valli ***@***.***> > > 发送时间:2020-07-15 >> 09:42:03 (星期三) > 收件人: ZitongYu/CDCN < > ***@***.***> > 抄送: >> xuhangxuhang ***@***.***>, > Mention < > >> ***@***.***> > 主题: Re: [ZitongYu/CDCN] Is > there >> overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could > you >> tell me which dataset you use to train? After training, how does > your > >> model performance at the data out of the dataset you train? > > I got good >> > results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test >> results > of CASIA-FASD and Replayattack is bad. > > can you please tell >> me, that how > did you use the Oulu dataset, because in > the dataset I >> found the videos. > > but in the source code they required the map_dir, >> even the train, test, > > dev required the images. > > — > You are >> receiving this because you were > mentioned. > Reply to this email >> directly, view it on GitHub, or > unsubscribe. > > — > You are receiving >> this because you commented. > Reply > to this email directly, view it on >> GitHub > < > >> #21 (comment)>, or > >> > unsubscribe > < > >> https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA > >> > > . > — You are receiving this because you were mentioned. Reply to this >> > email directly, view it on GitHub, or unsubscribe. > > — > You are >> receiving this because you commented. > Reply to this email directly, view >> it on GitHub > < >> #21 (comment)>, or > >> unsubscribe > < >> https://github.com/notifications/unsubscribe-auth/AMM7B2YW6ARN25YF3UIY4NDR5UKDBANCNFSM4N7N7IKA > >> > . > — You are receiving this because you were mentioned. Reply to this >> email directly, view it on GitHub, or unsubscribe. >> >> — >> You are receiving this because you commented. >> Reply to this email directly, view it on GitHub >> <#21 (comment)>, or >> unsubscribe >> < https://github.com/notifications/unsubscribe-auth/AMM7B22TINYGOLLCP62E6A3R5URRJANCNFSM4N7N7IKA > >> . >> > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#21 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMM7B25MM5T3VJRPBJJSALLR52YNRANCNFSM4N7N7IKA> .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

Answer 16 · 2020-07-29T02:58:18.000Z

my wechat id: spv_krv Can you please share your id? Thank you...

…

On Wed, 29 Jul 2020 at 10:54, xuhangxuhang ***@***.***> wrote: Sry I have no line app, I am from China, so you can connect me with Email or WeChat. -----原始邮件----- 发件人:punitha-valli ***@***.***> 发送时间:2020-07-29 10:35:30 (星期三) 收件人: ZitongYu/CDCN ***@***.***> 抄送: xuhangxuhang ***@***.***>, Mention < ***@***.***> 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) Thank you so much, Can i ask you a personal question? Can you share your line id, for future reference? My line id: spv.krv Thanks in advance On Tue, 28 Jul 2020 at 19:24, xuhangxuhang ***@***.***> wrote: > In Oulu-NPU dataset, I randoimly select 20 frames of each video of > dev/test subset, for training I balance data by video name first, for > example in Oulu-NPU Protocol-1 there are 240 real face video and 960 > spoofed ones if my memory is correct, I copy real samples for 4 times and > mixed the expaneded real video list and spoofed video list, shuffled them. > What I received is a balanced training set. Next randomly select one frame > from every balanced video, training them then. I set 10000 samples for an > epoch, and batchsize is 20. So For each epoch my model runs 500 steps. > That's it. > > > > > For your 2nd question, what you should do is just save the balanced > training name list as a .txt file, the same as Oulu-NPU dataset provided. > > > > > Sorry I can not show you my code now, I am already graduated, so my code > are deleted. > > > > > Any question, just ask me, I will replay you when I got time. > > > > > Best wish. > > > > > > > > -----原始邮件----- > 发件人:punitha-valli ***@***.***> > 发送时间:2020-07-27 16:25:04 (星期一) > 收件人: ZitongYu/CDCN ***@***.***> > 抄送: xuhangxuhang ***@***.***>, Mention < > ***@***.***> > 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) > > > > > > > Thank you for your kind response > > > How many frames did you used for one video ?, While converting video into > images. How did you saved the image names ?, I have some problem with this > . > > On Mon 27 Jul, 2020, 3:42 PM Punitha Valli, ***@***.***> > wrote: > > > Thank you for so much,. > > > > I have tried that one , but my dlib in not working properly. The wheel of > > dlib is not building correctly. > > > > > > On Mon 27 Jul, 2020, 3:07 PM xuhangxuhang, ***@***.***> > > wrote: > > > >> you can search PRNet iin GitHub, all map labels I have are generated > >> from that code. best wish hangxu ***@***.*** 签名由网易邮箱大师 > >> 定制 On 07/27/2020 14:34, punitha-valli wrote: Can you please share the > code > >> for making map_dir ? On Mon 27 Jul, 2020, 2:03 PM xuhangxuhang, < > >> ***@***.***> wrote: > What you should do is split the > title > >> video into single frames, for every > live frame yield depth map, for > each > >> spoof frame just yield an all zero > matrix, mapdir is the label save > >> folder. Is this clear? hangxu 邮箱： > ***@***.*** 签名由网易邮箱大师 > 定制 > >> On 07/27/2020 13:08, punitha-valli > wrote: Thank you for your response > I > >> have Oulu dataset, but it's fully > video. Then I don't understand about > >> the map_dir. Then the image path , > because I have videos in Oulu > >> dataset,.... What should I do for the > train_image, test_image, > dev_image, > >> map_dir, Can you please help me? Thank > you so much . On Sun 26 Jul, > 2020, > >> 11:18 AM xuhangxuhang, < > ***@***.***> wrote: > Hi，sorry > I > >> replay this email so late, > sincerely apologize for that. > > > > > In > >> Oulu-NPU dataset, the code > provider sent single image into their > > >> network. So what you should do is, > write a generator(in pytorch write > a > > >> class as the code in the Github > Repo, if you use TF or Keras, a > simple > > >> function is enough), the generator > should yeild image(face iamge) and > the > >> > corresponding depth map. In > training stage, get average output > score of > >> > each sample, in evaluation > stage, get prediction scores of samples > from > >> > original video and average > them as the final prediction result of > >> single > video. > > > > > (ps: I am > happy to reply your question, and > my > >> English is not that good, > if I > confuse you please ask again, I will > try > >> my best to answer.) > > Best wish. > > > > > > -----原始邮件----- > > >> 发件人:punitha-valli ***@***.***> > > 发送时间:2020-07-15 > >> 09:42:03 (星期三) > 收件人: ZitongYu/CDCN < > ***@***.***> > 抄送: > >> xuhangxuhang ***@***.***>, > Mention < > > >> ***@***.***> > 主题: Re: [ZitongYu/CDCN] Is > there > >> overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could > > you > >> tell me which dataset you use to train? After training, how does > your > > > >> model performance at the data out of the dataset you train? > > I got > good > >> > results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test > >> results > of CASIA-FASD and Replayattack is bad. > > can you please tell > >> me, that how > did you use the Oulu dataset, because in > the dataset I > >> found the videos. > > but in the source code they required the map_dir, > >> even the train, test, > > dev required the images. > > — > You are > >> receiving this because you were > mentioned. > Reply to this email > >> directly, view it on GitHub, or > unsubscribe. > > — > You are receiving > >> this because you commented. > Reply > to this email directly, view it on > >> GitHub > < > > >> #21 (comment)>, or > > > >> > unsubscribe > < > > >> > https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA > > > >> > > . > — You are receiving this because you were mentioned. Reply to > this > >> > email directly, view it on GitHub, or unsubscribe. > > — > You are > >> receiving this because you commented. > Reply to this email directly, > view > >> it on GitHub > < > >> #21 (comment)>, or > > > >> unsubscribe > < > >> > https://github.com/notifications/unsubscribe-auth/AMM7B2YW6ARN25YF3UIY4NDR5UKDBANCNFSM4N7N7IKA > > > >> > . > — You are receiving this because you were mentioned. Reply to this > >> email directly, view it on GitHub, or unsubscribe. > >> > >> — > >> You are receiving this because you commented. > >> Reply to this email directly, view it on GitHub > >> <#21 (comment)>, or > >> unsubscribe > >> < > https://github.com/notifications/unsubscribe-auth/AMM7B22TINYGOLLCP62E6A3R5URRJANCNFSM4N7N7IKA > > > >> . > >> > > > > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub, or unsubscribe. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#21 (comment)>, or > unsubscribe > < https://github.com/notifications/unsubscribe-auth/AMM7B25MM5T3VJRPBJJSALLR52YNRANCNFSM4N7N7IKA > > . > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#21 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMM7B27LXE2M6N3KF2ZVVCLR56FPXANCNFSM4N7N7IKA> .

Answer 17 · 2020-11-28T17:14:35.000Z

Hi，sorry I replay this email so late, sincerely apologize for that. In Oulu-NPU dataset, the code provider sent single image into their network. So what you should do is, write a generator(in pytorch write a class as the code in the Github Repo, if you use TF or Keras, a simple function is enough), the generator should yeild image(face iamge) and the corresponding depth map. In training stage, get average output score of each sample, in evaluation stage, get prediction scores of samples from original video and average them as the final prediction result of single video. (ps: I am happy to reply your question, and my English is not that good, if I confuse you please ask again, I will try my best to answer.) Best wish.
…
-----原始邮件----- 发件人:punitha-valli notifications@github.com 发送时间:2020-07-15 09:42:03 (星期三) 收件人: ZitongYu/CDCN CDCN@noreply.github.com 抄送: xuhangxuhang hangxu@my.swjtu.edu.cn, Mention mention@noreply.github.com 主题: Re: [ZitongYu/CDCN] Is there overfitting? (#21) @xuhangxuhang Thank you. Could you tell me which dataset you use to train? After training, how does your model performance at the data out of the dataset you train? I got good results inside Oulu, CASIA-FASD, and Replayattack, but cross-test results of CASIA-FASD and Replayattack is bad. can you please tell me, that how did you use the Oulu dataset, because in the dataset I found the videos. but in the source code they required the map_dir, even the train, test, dev required the images. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

you can search PRNet iin GitHub, all map labels I have are generated from that code. best wish hangxu 邮箱：hangxu@my.swjtu.edu.cn 签名由网易邮箱大师定制 On 07/27/2020 14:34, punitha-valli wrote: Can you please share the code for making map_dir ? On Mon 27 Jul, 2020, 2:03 PM xuhangxuhang, notifications@github.com wrote: > What you should do is split the title video into single frames, for every > live frame yield depth map, for each spoof frame just yield an all zero > matrix, mapdir is the label save folder. Is this clear? hangxu 邮箱： > hangxu@my.swjtu.edu.cn 签名由网易邮箱大师定制 On 07/27/2020 13:08, punitha-valli > wrote: Thank you for your response I have Oulu dataset, but it's fully > video. Then I don't understand about the map_dir. Then the image path , > because I have videos in Oulu dataset,.... What should I do for the > train_image, test_image, dev_image, map_dir, Can you please help me? Thank > you so much . On Sun 26 Jul, 2020, 11:18 AM xuhangxuhang, < > notifications@github.com> wrote: > Hi，sorry I replay this email so late, > sincerely apologize for that. > > > > > In Oulu-NPU dataset, the code > provider sent single image into their > network. So what you should do is, > write a generator(in pytorch write a > class as the code in the Github > Repo, if you use TF or Keras, a simple > function is enough), the generator > should yeild image(face iamge) and the > corresponding depth map. In > training stage, get average output score of > each sample, in evaluation > stage, get prediction scores of samples from > original video and average > them as the final prediction result of single > video. > > > > > (ps: I am > happy to reply your question, and my English is not that good, > if I > confuse you please ask again, I will try my best to answer.) > > Best wish. > > > > > > -----原始邮件----- > 发件人:punitha-valli notifications@github.com > > 发送时间:2020-07-15 09:42:03 (星期三) > 收件人: ZitongYu/CDCN < > CDCN@noreply.github.com> > 抄送: xuhangxuhang hangxu@my.swjtu.edu.cn, > Mention < > mention@noreply.github.com> > 主题: Re: [ZitongYu/CDCN] Is > there overfitting? (#21) > > > > > > > @xuhangxuhang Thank you. > > Could > you tell me which dataset you use to train? After training, how does > your > model performance at the data out of the dataset you train? > > I got good > results inside Oulu, CASIA-FASD, and Replayattack, but > cross-test results > of CASIA-FASD and Replayattack is bad. > > can you please tell me, that how > did you use the Oulu dataset, because in > the dataset I found the videos. > > but in the source code they required the map_dir, even the train, test, > > dev required the images. > > — > You are receiving this because you were > mentioned. > Reply to this email directly, view it on GitHub, or > unsubscribe. > > — > You are receiving this because you commented. > Reply > to this email directly, view it on GitHub > < > #21 (comment)>, or > > unsubscribe > < > https://github.com/notifications/unsubscribe-auth/AMM7B2YN3KOR4AATAVVIGKLR5ON75ANCNFSM4N7N7IKA> > > . > — You are receiving this because you were mentioned. Reply to this > email directly, view it on GitHub, or unsubscribe. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#21 (comment)>, or > unsubscribe > https://github.com/notifications/unsubscribe-auth/AMM7B2YW6ARN25YF3UIY4NDR5UKDBANCNFSM4N7N7IKA > . > — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

你好，我想请问一下包含bbox的.dat文件是怎样产生的呢，可以给一个相关网址吗，十分感谢

Answer 18 · 2021-09-19T04:51:22.000Z

Hi,

I also have the same problem with CelbA-spoofy data.

Is it a normal or abnormal case ? Because loss reduce very fast.

Answer 19 · 2021-11-16T02:58:40.000Z

Hi,

I also have the same problem with CelbA-spoofy data.

Is it a normal or abnormal case ? Because loss reduce very fast.

Sorry to bother you....could you tell me how to do the data preparation? especially the bbox dat file... thanks!

Answer 20 · 2021-11-17T02:11:14.000Z

Hi,
I also have the same problem with CelbA-spoofy data.
Is it a normal or abnormal case ? Because loss reduce very fast.

Sorry to bother you....could you tell me how to do the data preparation? especially the bbox dat file... thanks!

CelebA- Spoofing data used Retinaface for face detection. In the data, they propose available bounding box information, so if you want to follow them, you should read README.md for more information. Because, the bounding box information needs to convert x,y,h,w into real_x,real_y,real_h,real_w. In my opinion, I think that the dataset isn't clean, so you should be careful for data preprocessing. (more cross-check)