search

FVL2020 / ICCV-2023-MB-TaylorFormer

84 stars 9 forks source link

Supplementary material #2

Open hangxiaotian opened 11 months ago

hangxiaotian commented 11 months ago

Can you share the supplementary material?

raindrop313 commented 11 months ago

Yes, please give me your e-mail.

duanpingtin commented 11 months ago

Can you share the supplementary material and your datasets for train and test?my e-mail is 1563565613@qq.com.

xinxing277 commented 11 months ago

Can you share the supplementary material and your datasets for train and test?my e-mail is 1272007308@qq.com. Thanks!

raindrop313 commented 10 months ago

@duanpingtin @xinxing277 I'll upload the dataset to Google Cloud Drive and send it to you later.

duanpingtin commented 10 months ago

What are the requirements for computing power? Two 3090s can’t run it.

raindrop313 commented 10 months ago

@duanpingtin Please provide more information, e.g., which model was used, the batch_size of each 3090 and the size of the images

duanpingtin commented 10 months ago

@raindrop313 MB-TaylorFormer-B.yml size of the images is 640*640, I have not changed the batch_size size of the original code

raindrop313 commented 10 months ago

@duanpingtin A 640640 image as training input is so large that the 3090 can't run it. You can set the size of the random crop in the "option" file to 256256 with batch_size set to 4, or 512*512 with batch_size set to 1.

raindrop313 commented 10 months ago

@duanpingtin @xinxing277 @hangxiaotian Download links for supplementary materials and datasets have been updated in the “readme”