search

xiaomengyc / ACoL

(CVPR2018) Adversarial Complementary Learning for Weakly Supervised Object Localization
http://openaccess.thecvf.com/content_cvpr_2018/papers/Zhang_Adversarial_Complementary_Learning_CVPR_2018_paper.pdf
MIT License
164 stars 30 forks source link

picture preprocessing problem #3

Open tim120526 opened 5 years ago

tim120526 commented 5 years ago

Hi! @xiaomengyc I saw the images were resized to 256x256 then randomly cropped to 224x224 during training,but in the code ,smaller edge of the image will be resized to 256,then randomly cropped. emm..i desire to know how to resize the images during training and test,especially for the localization test. Thank you very much!Good luck to you!

xiaomengyc commented 5 years ago

@tim20120526 Images are firstly resized to 256256, and then cropped to 224224 in the training stage. For the testing stage, images are directly resized to 224*224. Sorry, I didn't get your point of resizing the smaller edge to 256. I think this is the common technique adopted in detection tasks rather than classification tasks. Best.