Hi,
By inspecting the image Flooded image 7289.jpg from track-1
(`FloodNet Challenge @ EARTHVISION 2021 - Track 1/Train/Labeled/Flooded/image/7289.jpg`)
and the Non-Flooded image 6692.jpg also from track-1
(`FloodNet Challenge @ EARTHVISION 2021 - Track 1/Train/Labeled/Non-Flooded/image/6629.jpg`)
it is clear that they depict the same place, just by varying a little bit the angle of the viewpoint of the camera. It seems that the left part of image 7289.jpg is what makes it be classified as Flooded. Nevertheless, by inspecting the labels in its corresponding mask there are no pixels with the labels `1 Building-flooded` and there are also no pixels labeled as `3 Road-flooded`.
Is this a mistake, there is some degree of subjectiveness to decide if an image like 7289.jpg should be classified as Flooded, or maybe there are some other rules?
I've posted the same question at the repository in github with the images: https://github.com/BinaLab/FloodNet-Challenge-EARTHVISION2021/issues/3
Thanks in advance,
Artur
When most of the part of an image is covered in flood water, the image is classified as Flooded, otherwise Non-flooded. But there are some degree of subjectivities.
Thank you!
Posted by: binalab @ April 18, 2021, 4:08 a.m.