search

xuebinqin / U-2-Net

The code for our newly accepted paper in Pattern Recognition 2020: "U^2-Net: Going Deeper with Nested U-Structure for Salient Object Detection."
Apache License 2.0
8.52k stars 1.47k forks source link

Use Salient Object Detection to realize Salient-Instance Segmentation #11

Closed lfxx closed 4 years ago

lfxx commented 4 years ago

Is it possible to realize this task?@NathanUA

Adrian147 commented 4 years ago

Hi, these could be helpful to you from their previous paper:

  1. https://github.com/NathanUA/BASNet/issues/30
  2. https://github.com/NathanUA/BASNet/issues/21
xuebinqin commented 4 years ago

Thanks for your question and Adrian147's answer. We were planing to try semantic segmentation based on our model. But that requires more computation resources and then we gave it up. We hope that U2Net to be a baseline for binary segmentation not necessarily for salient object detection (actually we tried different tasks e.g. building extraction, medical image segmentation, etc, u2net also give very good results on these tasks.) As for the multiple classes segmentation, we haven't try yet. So we are not able to claim the performance. But we believe it is promising in semantic segmentation and we may try that later. If you are interested, pls feel free to give it a try. To use it for semantic segmentation, you have to change the last layer from sigmoid to softmax, change to the bce loss to ce loss, and modify the dataloader to read multiple class labels.

lfxx commented 4 years ago

I will try it.Thanks for your rapidly reply:)

lfxx commented 4 years ago

Thanks for your question and Adrian147's answer. We were planing to try semantic segmentation based on our model. But that requires more computation resources and then we gave it up. We hope that U2Net to be a baseline for binary segmentation not necessarily for salient object detection (actually we tried different tasks e.g. building extraction, medical image segmentation, etc, u2net also give very good results on these tasks.) As for the multiple classes segmentation, we haven't try yet. So we are not able to claim the performance. But we believe it is promising in semantic segmentation and we may try that later. If you are interested, pls feel free to give it a try. To use it for semantic segmentation, you have to change the last layer from sigmoid to softmax, change to the bce loss to ce loss, and modify the dataloader to read multiple class labels.

There may be a misunderstood.I want to realize the Salient-Instance Segmentation which similiar to S4Net not purely instance segmentation.What should i do with u2net?

xuebinqin commented 4 years ago

There are mainly two ways for instance segmentation. (1) add another network or (decoder of the u2net) to regress the bounding boxes, each bounding box represents an instance. The segmentation will happen within each bounding box. (2) another way is the metric learning e.g. https://arxiv.org/pdf/1703.10277.pdf.

On Tue, May 12, 2020 at 8:22 PM 高峰 notifications@github.com wrote:

Reopened #11 https://github.com/NathanUA/U-2-Net/issues/11.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/NathanUA/U-2-Net/issues/11#event-3330035043, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORKINNBQ36MG7TFZ3TDRRH77VANCNFSM4M6T7UYA .

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

lfxx commented 4 years ago

There are mainly two ways for instance segmentation. (1) add another network or (decoder of the u2net) to regress the bounding boxes, each bounding box represents an instance. The segmentation will happen within each bounding box. (2) another way is the metric learning e.g. https://arxiv.org/pdf/1703.10277.pdf. … On Tue, May 12, 2020 at 8:22 PM 高峰 @.***> wrote: Reopened #11 <#11>. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#11 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADSGORKINNBQ36MG7TFZ3TDRRH77VANCNFSM4M6T7UYA . -- Xuebin Qin PhD Department of Computing Science University of Alberta, Edmonton, AB, Canada Homepage:https://webdocs.cs.ualberta.ca/~xuebin/

Get it.Thanks very much