How to get data augmentation datasets?

[xuebinqin]

You can download the DUTS-TR from http://saliencydetection.net/duts/ and conduct horizontal flipping on the training set to double the training images for augmentation.

On Sun, Jun 2, 2019 at 6:05 AM panmyuan notifications@github.com wrote:

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/NathanUA/BASNet/issues/4?email_source=notifications&email_token=ADSGORKNKXF6WVCGT6TYXRTPYOZPFA5CNFSM4HSCP4M2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4GXEQ4YQ, or mute the thread https://github.com/notifications/unsubscribe-auth/ADSGOROBJNQYI7ZS5VGEDYLPYOZPFANCNFSM4HSCP4MQ .

-- Xuebin Qin PhD Candidate Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

[panmyuan]

ok,I have solved this problem. Thank you very much!

[panmyuan]

I have problems about saving training models.A batch size I set is 12, how can I set how many iterations to save the model?
I see your model saving the model every iteration, so some models are meaningless.

[xuebinqin]

Try to change the line 169 to "if ite_num % 2000 == 0:" which means you are going to save the model every 2000 iterations.

On Sun, Jun 2, 2019 at 9:09 PM panmyuan notifications@github.com wrote:

I have problems about saving training models.A batch size I set is 12, how can I set how many iterations to save the model? I see your model saving the model every iteration, so some models are meaningless.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/NathanUA/BASNet/issues/4?email_source=notifications&email_token=ADSGORML2CPBCA2KWTC47XTPYSDOBA5CNFSM4HSCP4M2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWYGBEQ#issuecomment-498098322, or mute the thread https://github.com/notifications/unsubscribe-auth/ADSGORMQ5K5ETQXFJUISDCTPYSDOBANCNFSM4HSCP4MQ .

-- Xuebin Qin PhD Candidate Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

[panmyuan]

I think saving the model every 2000 iterations is equivalent to 1 epoch to save the model. But total epochs are 100000. It is so large and those saved models will take up a lot of storage.

[xuebinqin]

You can reduce the total epochs or increase the saving intervals to a smaller one. Actually, I stopped the training process after 400K iterations.

On Sun, Jun 2, 2019 at 9:58 PM panmyuan notifications@github.com wrote:

I think saving the model every 2000 iterations is equivalent to 1 epoch to save the model. But total epochs are 100000. It is so large and those saved models will take up a lot of storage.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/NathanUA/BASNet/issues/4?email_source=notifications&email_token=ADSGORMAMTXSP4UB72F6STDPYSJGJA5CNFSM4HSCP4M2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWYHUMY#issuecomment-498104883, or mute the thread https://github.com/notifications/unsubscribe-auth/ADSGORLQN4RNGTOWT4OQEXLPYSJGJANCNFSM4HSCP4MQ .

-- Xuebin Qin PhD Candidate Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

[xuebinqin]

You can also try to keep the best 5 or 10 models based on the training loss.

On Sun, Jun 2, 2019 at 10:01 PM Xuebin Qin xuebin@ualberta.ca wrote:

You can reduce the total epochs or increase the saving intervals to a smaller one. Actually, I stopped the training process after 400K iterations.

On Sun, Jun 2, 2019 at 9:58 PM panmyuan notifications@github.com wrote:

I think saving the model every 2000 iterations is equivalent to 1 epoch to save the model. But total epochs are 100000. It is so large and those saved models will take up a lot of storage.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/NathanUA/BASNet/issues/4?email_source=notifications&email_token=ADSGORMAMTXSP4UB72F6STDPYSJGJA5CNFSM4HSCP4M2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWYHUMY#issuecomment-498104883, or mute the thread https://github.com/notifications/unsubscribe-auth/ADSGORLQN4RNGTOWT4OQEXLPYSJGJANCNFSM4HSCP4MQ .

-- Xuebin Qin PhD Candidate Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

-- Xuebin Qin PhD Candidate Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

[panmyuan]

ok, I know.As you said, I guess you stopped training around 150 epochs.

[xuebinqin]

Yes, exactly.

On Sun, Jun 2, 2019 at 10:11 PM panmyuan notifications@github.com wrote:

ok, I know.As you said, I guess you stopped training around 150 epochs.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/NathanUA/BASNet/issues/4?email_source=notifications&email_token=ADSGORJI3UVQLRFG5QYB75TPYSKVTA5CNFSM4HSCP4M2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWYIBIQ#issuecomment-498106530, or mute the thread https://github.com/notifications/unsubscribe-auth/ADSGOROW3X7JRX4Z3BTXFU3PYSKVTANCNFSM4HSCP4MQ .

-- Xuebin Qin PhD Candidate Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/