search

houqb / SeeNet

Self-Erasing Network for Integral Object Attention
54 stars 9 forks source link

segmentation model #1

Closed tim120526 closed 5 years ago

tim120526 commented 5 years ago

Hi! @Andrew-Qibin , Thanks for sharing your code! I appreciate the Erase method! I see you use Deeplab for segmentation task in paper. I am a bit confused, are you use deeplabv2_ASPP to implement segmentation task based on vgg16 and resnet101 backbone, i see your paper cite deeplabv2 paper but use Deeplab-LargeFOV. emm.. maybe you use Deeplab-LargeFOV based on vgg16 and use deeplabv2_ASPP based on resnet101? I am confused. Can u give me some ides? Thank you very much!

houqb commented 5 years ago

I just use Deeplab-LargeFOV with ResNet-101. You can also use deeplabv2 with ASPP and let others know what you use. I empirically found Deeplabv2 with ASPP is not that helpful as the version I used.

tim120526 commented 5 years ago

Get it. Thanks for your advice.

--------------原始邮件-------------- 发件人："Qibin (Andrew) Hou "notifications@github.com; 发送时间：2018年11月27日(星期二) 晚上9:58 收件人："Andrew-Qibin/SeeNet" SeeNet@noreply.github.com; 抄送："tim20120526 "2111704114@mail2.gdut.edu.cn;"Author "author@noreply.github.com; 主题：Re: [Andrew-Qibin/SeeNet] segmentation model (#1)

I just use Deeplab-LargeFOV with ResNet-101. You can also use deeplabv2 with ASPP and let others know what you use. I empirically found Deeplabv2 with ASPP is not that helpful as the version I used.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.