search

recasens / Gaze-Following

Code for Gaze-Following in video
http://videogazefollow.csail.mit.edu
47 stars 6 forks source link

training time #3

Open xudanni0927 opened 4 years ago

xudanni0927 commented 4 years ago

I am trying to train this model, when I set epoches value to 1, it takes several hours to train and now I am trying to set epoches value to 900. I'd like to know how long it took you to train the model to achieve good results, and what is the value of epoches.
Thanks!

le-wei commented 4 years ago

@xudanni0927 Hello, I encountered problems with train.txt and test.txt when using train.py, mainly because I don't understand where these data come from. Would you like to share a copy of your train.txt and test.txt for me to learn, thank you very much. My address is ‘lv521lup@gmail.com’. Thanks again

xudanni0927 commented 4 years ago

@xudanni0927 Hello, I encountered problems with train.txt and test.txt when using train.py, mainly because I don't understand where these data come from. Would you like to share a copy of your train.txt and test.txt for me to learn, thank you very much. My address is ‘lv521lup@gmail.com’. Thanks again

Sorry, I just noticed this message till now, do you need the training and testing files now?

le-wei commented 4 years ago

@xudanni0927 Yes, I still need training and testing files. Thank you very much for taking the time to share these files. My address is lv521lup@gmail.com

xudanni0927 commented 4 years ago

Hi， These are the files I generated from the Videogaze dataset. Hoping these helps.

Gracie Xu

------------------ 原始邮件 ------------------ 发件人: "recasens/Gaze-Following" <notifications@github.com>; 发送时间: 2020年7月14日(星期二) 上午10:16 收件人: "recasens/Gaze-Following"<Gaze-Following@noreply.github.com>; 抄送: "1770774149"<1770774149@qq.com>;"Mention"<mention@noreply.github.com>; 主题: Re: [recasens/Gaze-Following] training time (#3)

@xudanni0927 Yes, I still need training and testing files. Thank you very much for taking the time to share these files. My address is lv521lup@gmail.com

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

从QQ邮箱发来的超大附件

test_flag.txt (6.43M, 2020年07月14日 10:22 到期)进入下载页面：http://mail.qq.com/cgi-bin/ftnExs_download?t=exs_ftn_download&k=77323033961dfc9cb0ecbb5a1339574c5301060b005a0052180052060d14510253531d0b025a554e0d535252055c525b07055200353465175041446c535504041b4648473504&code=520359ec

train_flag.txt (72.58M, 2020年08月13日 10:22 到期)进入下载页面：http://mail.qq.com/cgi-bin/ftnExs_download?t=exs_ftn_download&k=7065326249da179db6bbb90b13635349565d0a52040751561e570055054e55000b571f000c53044b500450570301595602035301356d611241045b0c6a050d07544b461a41635c&code=3e2b5caf

le-wei commented 4 years ago

@xudanni0927 This is exactly what I need, thank you very much for your help。

xudanni0927 commented 4 years ago

Hi, I just find out that the eye position need to be normalized in the range [0.0, 1.0]. You need to normalize it before your training and testing.

Wish you best.

------------------ 原始邮件 ------------------ 发件人: "recasens/Gaze-Following" <notifications@github.com>; 发送时间: 2020年7月14日(星期二) 晚上7:11 收件人: "recasens/Gaze-Following"<Gaze-Following@noreply.github.com>; 抄送: "1770774149"<1770774149@qq.com>;"Mention"<mention@noreply.github.com>; 主题: Re: [recasens/Gaze-Following] training time (#3)

@xudanni0927 This is exactly what I need, thank you very much for your help。

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

le-wei commented 4 years ago

@xudanni0927 Hello . I will standardize the position of the eyes before training, thank you very much for your guidance. good luck

xudanni0927 commented 4 years ago

Hi, Sorry to bother you. I am learning to train this model, too.  I found another problem in these files I generated, I will send you the updated ones to your email.  Best wishes.

------------------ 原始邮件 ------------------ 发件人: "recasens/Gaze-Following" <notifications@github.com>; 发送时间: 2020年7月14日(星期二) 晚上7:41 收件人: "recasens/Gaze-Following"<Gaze-Following@noreply.github.com>; 抄送: "1770774149"<1770774149@qq.com>;"Mention"<mention@noreply.github.com>; 主题: Re: [recasens/Gaze-Following] training time (#3)

@xudanni0927 Hello . I will standardize the position of the eyes before training, thank you very much for your guidance. good luck

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

le-wei commented 4 years ago

@xudanni0927 I will use your updated file. I believe that with your help I can train this model successfully. Thank you very much for your warm help. Wish you happy every day, always beautiful.

xudanni0927 commented 4 years ago

@xudanni0927 @liweilup Hi, I can't train my own model because of the same problem. Could you please share a copy of train.txt and test.txt?

My e-mail is bluence75@gmail.com

The train_flag.txt and test_flag.txt are generated from the videogaze.mat file. Hope them help. I used the test_flag.txt to evaluate the model and it worked out well. Please note that the order of target-source-face file paths in train_flag.txt and test_flag.txt should to be consistent with the reading order in dataloader.py.

xuanzhao commented 3 years ago

Hello, I also need the dataset to run the code, could you pls share a copy of train_flag.txt and test_flag.txt? my e-mail is xuanzhao89@163.com @xudanni0927 @liweilup @BLUENCE

hisung97 commented 3 years ago

Hello, I also need the dataset to run the code, could you please share a copy of train_flag.txt and test_flag.txt? my e-mail is hisung6759@naver.com @xudanni0927 @liweilup @BLUENCE

lfy2000 commented 2 years ago

@xudanni0927@liweilup嗨，由于同样的问题，我无法训练自己的模型。你能分享一份火车.txt和测试.txt的副本吗？ 我的电子邮件已 bluence75@gmail.com

train_flag.txt和test_flag.txt是从视频增益仪.mat 文件生成的。希望他们有所帮助。我使用test_flag.txt来评估模型，效果很好。请注意，train_flag.txt和test_flag.txt中目标源面文件路径的顺序应与 dataloader.py 中的读取顺序一致。

Can you still find these two txt files because the files I generated seem to have problems

csuwangchengqiao commented 1 year ago

hello，guys, I am eager to run this code. Could you send me this train_flag.txt and test_flag.txt to my Email: 350453547@qq.com, thank you in adcance.@liweilup @xudanni0927 @hisung97 @lfy2000

Sponge-bo commented 1 year ago

hi，guys，I would very much like the train_flag.txt and test_flag.txt for this code,Could you send me this train_flag.txt and test_flag.txt to my Email: 1765695559@qq.com,thank a lot! @liweilup @xudanni0927 @hisung97 @csuwangchengqiao @lfy2000

MiaoHN commented 4 months ago

Hello, everyone!

I generated train_flag.txt and test_flag.txt from the data in videogaze.mat. Here are the statistics:

File Size Lines
train_flag.txt 36M 133,565
test_flag.txt 11M 40,642

Can you confirm if this is correct?

I ran the model, but the results were not very good, with an L2-Distance metric of only 0.241. Has anyone encountered this issue?

PcCin37 commented 3 weeks ago

@xudanni0927 @liweilup Hi, I can't train my own model because of the same problem. Could you please share a copy of train.txt and test.txt? My e-mail is bluence75@gmail.com

The train_flag.txt and test_flag.txt are generated from the videogaze.mat file. Hope them help. I used the test_flag.txt to evaluate the model and it worked out well. Please note that the order of target-source-face file paths in train_flag.txt and test_flag.txt should to be consistent with the reading order in dataloader.py.

I would very much like the train_flag.txt and test_flag.txt for this code,Could you send me this train_flag.txt and test_flag.txt to my Email: pengchenjw@163.com,thank a lot! @xudanni0927