search

suyukun666 / S2CNet

Official PyTorch implementation of the “Spatial-Semantic Collaborative Cropping for User Generated Content”. (AAAI24)
42 stars 4 forks source link

is is a bug when flip the image? #12

Open dongdk opened 1 month ago

dongdk commented 1 month ago

hi @suyukun666 i wonder these two lines are bug? since flip the image, it should flip the x1, x2 correspondingly to make sure x2 >= x1. when i try to fix it like this, and train a model, the performance gets worse.

                tmp = annotations[i][1]
                annotations[i][1] = width - annotations[i][3]
                annotations[i][3] = width - tmp