Cropping's affect on annatations?

Tencent / tencent-ml-images

Largest multi-label image database; ResNet-101 model; 80.73% top-1 acc on ImageNet

Other

3.05k stars 514 forks source link

Cropping's affect on annatations? #64

Open gao-shi opened 4 years ago

gao-shi commented 4 years ago

In the paper, it is metioned that random crop a bounding box from input image, for which the box area is within [0.05, 1.0] of the whole image area for image pre-processing. But how can i know which label should be kept? It could happen that the object is cut off，but its label remains in the gt annotations.

yunkchen commented 4 years ago

Params in train.py: object_cover=0.7, area_cover=0.7, I guess that just to keep most images' label invariant.