search

jingyuanli001 / RFR-Inpainting

The source code for CVPR 2020 accepted paper "Recurrent Feature Reasoning for Image Inpainting"
MIT License
355 stars 76 forks source link

About trainging masks dataset #30

Closed haolin512900 closed 2 years ago

haolin512900 commented 3 years ago

Hi，Thank you for your great project again！ How do you divide the masks training dataset from pconv? Whether to use all masks training data sets as training data sets ?

haolin512900 commented 3 years ago

Or you train your model using mask_mode==1？

jingyuanli001 commented 3 years ago

For the training mask in pconv's paper, you might first randomly dilate the holes to enlarge the masked region (this could be done with max pooling on binary mask). Then you should do random cropping, flipping and resizing to augment the masks. For the mask preparation details, you might refer to the procedure released by Edge-Connect. You can also use mask_mode 1 to train the model to inpaint with random masks.

haolin512900 commented 3 years ago

Thank your reply! good lucky

haolin512900 commented 3 years ago

Thank your reply! Good lucky

------------------ 原始邮件 ------------------ 发件人: "jingyuanli001/RFR-Inpainting" <notifications@github.com>; 发送时间: 2020年12月24日(星期四) 晚上8:42 收件人: "jingyuanli001/RFR-Inpainting"<RFR-Inpainting@noreply.github.com>; 抄送: "1250404756"<1250404756@qq.com>;"Author"<author@noreply.github.com>; 主题: Re: [jingyuanli001/RFR-Inpainting] About trainging masks dataset (#30)

For the training mask in pconv's paper, you might first randomly dilate the holes to enlarge the masked region (this could be done with max pooling on binary mask). Then you should do random cropping, flipping and resizing to augment the masks. For the mask preparation details, you might refer to the procedure released by Edge-Connect. You can also use mask_mode 1 to train the model to inpaint with random masks.

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