mask post-processing

[xiongzhu666]

Hi Xuebin, I implement a mask post-processing by c++ to retain the segmentation detail, but perhaps it only has good performance on sky-segmentation。thanks for your kind help in my model training. https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing

[xuebinqin]

Hi, xiong,

Thanks for your interests and awesome works. Your dataset and results look really impressive. BTW, do you mind that I include some of your figures and your github link in the readme of our U^2-Net repo ? I think it probably good for advertising your product and repo. In addition, our next version U^2-Net and a new great dataset are almost there. We will release them in the near future. If you are interested, please be aware of our updates.

All the best, Xuebin

On Wed, Oct 27, 2021 at 7:04 AM xiongzhu666 @.***> wrote:

Hi Xuebin, I implement a mask post-processing by c++ to retain the segmentation detail, but perhaps it only has good performance on sky-segmentation。thanks for your kind help in my model training. https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/xuebinqin/U-2-Net/issues/262, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORK3C7547NZJWU6LOBTUI5T47ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

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

[xiongzhu666]

@xuebinqin Hi Xuebin, it is my pleasure that you include my repo in yours

1. Yes, the datasets are very important for segmentation, I just relabeled the 9000 ADE20K datasets and add 4000 private datasets from company. but it isn't still good enough to generate high-resolution mask. because the input resolution is small(320-448) considering speed, So I do the mask post-processing by another paper published by Google. But that paper is only for sky mask post-processing. so perhaps It won't get the same good performance on other classes segmentation.

2. I will continue to pay attention to your work

[xiongzhu666]

Hi, xiong, Thanks for your interests and awesome works. Your dataset and results look really impressive. BTW, do you mind that I include some of your figures and your github link in the readme of our U^2-Net repo ? I think it probably good for advertising your product and repo. In addition, our next version U^2-Net and a new great dataset are almost there. We will release them in the near future. If you are interested, please be aware of our updates. All the best, Xuebin … On Wed, Oct 27, 2021 at 7:04 AM xiongzhu666 @.***> wrote: Hi Xuebin, I implement a mask post-processing by c++ to retain the segmentation detail, but perhaps it only has good performance on sky-segmentation。thanks for your kind help in my model training. https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#262>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORK3C7547NZJWU6LOBTUI5T47ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub. -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

some funny application, : )

[xuebinqin]

Looks really impressive. Since I am currently busy with our newer version model and dataset, I will add your awesome results in the readme in the next few weeks. I think your works will be inspiring to many people in different fields.

Best regards,

On Fri, Oct 29, 2021 at 6:53 AM xiongzhu666 @.***> wrote:

Hi, xiong, Thanks for your interests and awesome works. Your dataset and results look really impressive. BTW, do you mind that I include some of your figures and your github link in the readme of our U^2-Net repo ? I think it probably good for advertising your product and repo. In addition, our next version U^2-Net and a new great dataset are almost there. We will release them in the near future. If you are interested, please be aware of our updates. All the best, Xuebin … <#m_-347751780870611050_> On Wed, Oct 27, 2021 at 7:04 AM xiongzhu666 @.***> wrote: Hi Xuebin, I implement a mask post-processing by c++ to retain the segmentation detail, but perhaps it only has good performance on sky-segmentation。thanks for your kind help in my model training. https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#262 https://github.com/xuebinqin/U-2-Net/issues/262>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORK3C7547NZJWU6LOBTUI5T47ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

some funny application, : ) [image: fun1] https://user-images.githubusercontent.com/52955431/139365803-faf02744-bd55-4f30-bbd3-9203f2bfd301.png

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/xuebinqin/U-2-Net/issues/262#issuecomment-954367012, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORIC3F4NTBWKZUIRULTUJIEB5ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

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

[xiongzhu666]

Looks really impressive. Since I am currently busy with our newer version model and dataset, I will add your awesome results in the readme in the next few weeks. I think your works will be inspiring to many people in different fields. Best regards, On Fri, Oct 29, 2021 at 6:53 AM xiongzhu666 @.> wrote: … Hi, xiong, Thanks for your interests and awesome works. Your dataset and results look really impressive. BTW, do you mind that I include some of your figures and your github link in the readme of our U^2-Net repo ? I think it probably good for advertising your product and repo. In addition, our next version U^2-Net and a new great dataset are almost there. We will release them in the near future. If you are interested, please be aware of our updates. All the best, Xuebin … <#m_-347751780870611050_> On Wed, Oct 27, 2021 at 7:04 AM xiongzhu666 @.> wrote: Hi Xuebin, I implement a mask post-processing by c++ to retain the segmentation detail, but perhaps it only has good performance on sky-segmentation。thanks for your kind help in my model training. https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#262 <#262>>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORK3C7547NZJWU6LOBTUI5T47ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/ some funny application, : ) [image: fun1] https://user-images.githubusercontent.com/52955431/139365803-faf02744-bd55-4f30-bbd3-9203f2bfd301.png — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#262 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORIC3F4NTBWKZUIRULTUJIEB5ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub. -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

Hi Xuebin, wait for your new version. in addition, I have some question to ask you:

1. the u2netp is much smaller than u2net, but the interence speed doesn't increase obviously. the reason is that the struct of network is nested？？？
2. Can you recommend any other high-speed networks ( for example Mobilenetv2+Unet??)，I want it can interence in real-time in mobile device.

Best regards

[xuebinqin]

Hi, Xiongzhu,

Sorry for the late reply. We were very busy with the cvpr22 submission in the past weeks. We just got time for updating our Readme file. Your repo for sky segmentation has been listed there. https://github.com/xuebinqin/U-2-Net/blob/master/README.md

1. the u2netp is much smaller than u2net, but the interence speed doesn't increase obviously. the reason is that the struct of network is nested？？？
2. Can you recommend any other high-speed networks ( for example Mobilenetv2+Unet??)，I want it can interence in real-time in mobile device. RE: You can give it a try. The bottleneck of U^2-Net and U^2-Net is not from the model size, but the depth. Both our models have very deep architectures, which impedes the real-time performance on some of the hardwares. But some other researchers and developes converted them into ONNX or tensorlithe and then they are able to run on CPU and cellphones. I am not quite sure about the conversion. You can google it. Shouldn't be that complicated. As for the real-time segmentation models, this repo https://github.com/YuvalNirkin/hyperseg could be a possible alternative(CVPR2021). You can also find some counterparts from their paper.

Best of luck,

On Sat, Oct 30, 2021 at 8:54 AM xiongzhu666 @.***> wrote:

Looks really impressive. Since I am currently busy with our newer version model and dataset, I will add your awesome results in the readme in the next few weeks. I think your works will be inspiring to many people in different fields. Best regards, On Fri, Oct 29, 2021 at 6:53 AM xiongzhu666 @.

> wrote: … <#m_4808286652383061102_> Hi, xiong, Thanks for your interests and awesome works. Your dataset and results look really impressive. BTW, do you mind that I include some of your figures and your github link in the readme of our U^2-Net repo ? I think it probably good for advertising your product and repo. In addition, our next version U^2-Net and a new great dataset are almost there. We will release them in the near future. If you are interested, please be aware of our updates. All the best, Xuebin … <#m_-347751780870611050_> On Wed, Oct 27, 2021 at 7:04 AM xiongzhu666 @.> wrote: Hi Xuebin, I implement a mask post-processing by c++ to retain the segmentation detail, but perhaps it only has good performance on sky-segmentation。thanks for your kind help in my model training. https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#262 https://github.com/xuebinqin/U-2-Net/issues/262 <#262 https://github.com/xuebinqin/U-2-Net/issues/262>>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORK3C7547NZJWU6LOBTUI5T47ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/ some funny application, : ) [image: fun1] https://user-images.githubusercontent.com/52955431/139365803-faf02744-bd55-4f30-bbd3-9203f2bfd301.png — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#262 (comment) https://github.com/xuebinqin/U-2-Net/issues/262#issuecomment-954367012>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORIC3F4NTBWKZUIRULTUJIEB5ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

Hi Xuebin, wait for your new version. in addition, I have some question to ask you:

1. the u2netp is much smaller than u2net, but the interence speed doesn't increase obviously. the reason is that the struct of network is nested？？？
2. Can you recommend any other high-speed networks ( for example Mobilenetv2+Unet??)，I want it can interence in real-time in mobile device.

Best regards

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/xuebinqin/U-2-Net/issues/262#issuecomment-955148155, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORLOAK5SP3EHGANCFZDUJN3ARANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

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

[xiongzhu666]

Hi, Xiongzhu, Sorry for the late reply. We were very busy with the cvpr22 submission in the past weeks. We just got time for updating our Readme file. Your repo for sky segmentation has been listed there. https://github.com/xuebinqin/U-2-Net/blob/master/README.md 1. the u2netp is much smaller than u2net, but the interence speed doesn't increase obviously. the reason is that the struct of network is nested？？？ 2. Can you recommend any other high-speed networks ( for example Mobilenetv2+Unet??)，I want it can interence in real-time in mobile device. RE: You can give it a try. The bottleneck of U^2-Net and U^2-Net is not from the model size, but the depth. Both our models have very deep architectures, which impedes the real-time performance on some of the hardwares. But some other researchers and developes converted them into ONNX or tensorlithe and then they are able to run on CPU and cellphones. I am not quite sure about the conversion. You can google it. Shouldn't be that complicated. As for the real-time segmentation models, this repo https://github.com/YuvalNirkin/hyperseg could be a possible alternative(CVPR2021). You can also find some counterparts from their paper. Best of luck, On Sat, Oct 30, 2021 at 8:54 AM xiongzhu666 @.***> wrote: … Looks really impressive. Since I am currently busy with our newer version model and dataset, I will add your awesome results in the readme in the next few weeks. I think your works will be inspiring to many people in different fields. Best regards, On Fri, Oct 29, 2021 at 6:53 AM xiongzhu666 @. > wrote: … <#m_4808286652383061102_> Hi, xiong, Thanks for your interests and awesome works. Your dataset and results look really impressive. BTW, do you mind that I include some of your figures and your github link in the readme of our U^2-Net repo ? I think it probably good for advertising your product and repo. In addition, our next version U^2-Net and a new great dataset are almost there. We will release them in the near future. If you are interested, please be aware of our updates. All the best, Xuebin … <#m_-347751780870611050_> On Wed, Oct 27, 2021 at 7:04 AM xiongzhu666 @.> wrote: Hi Xuebin, I implement a mask post-processing by c++ to retain the segmentation detail, but perhaps it only has good performance on sky-segmentation。thanks for your kind help in my model training. https://github.com/xiongzhu666/Sky-Segmentation-and-Post-processing — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#262 <#262> <#262 <#262>>>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORK3C7547NZJWU6LOBTUI5T47ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/ some funny application, : ) [image: fun1] https://user-images.githubusercontent.com/52955431/139365803-faf02744-bd55-4f30-bbd3-9203f2bfd301.png — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#262 (comment) <#262 (comment)>>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORIC3F4NTBWKZUIRULTUJIEB5ANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub . -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/ Hi Xuebin, wait for your new version. in addition, I have some question to ask you: 1. the u2netp is much smaller than u2net, but the interence speed doesn't increase obviously. the reason is that the struct of network is nested？？？ 2. Can you recommend any other high-speed networks ( for example Mobilenetv2+Unet??)，I want it can interence in real-time in mobile device. Best regards — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#262 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORLOAK5SP3EHGANCFZDUJN3ARANCNFSM5GZH4P7Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub. -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

Xuebin, really thanks for your kind suggestion and I am very concerned and look forward to your future work :) best regards