I have downloaded the dataset and i find 3 missing classes: riders, motos and trains. Furthermore, the class bike is only in around 30 images and those 30 images contain around 40 pixels of class bike.
Is there a mistake with the published dataset?
In the data description it's mentioned that this data support 4 tasks which is object detection, instance segmentation and semantic segmentation. I can find the labels of 2D bounding box for object detection.
Here I can't find the 2D bounding box
could you please add the 2D bounding box for object detection if you guys do have it?
I have read the paper and this statement in the paper indicate that model trained on MUAD images modified with simple histogram matching with Cityscapes images can achieve better results than w/o histogram matching.
I am curious about how to choose the reference image for histogram matching. Is the reference image randomly chosen from the target dataset(Cityscapes)?
Hello !
In the readme, classes are indexed from 0 to 20. When reading the files 000***_leftLabel.png i find some pixel value at 255. What does it mean ?
There is also no busses in the train dataset (idx 16) even though this class is not supposed to be OOD. Can you explain why ?
In the expected output should be to predict classes up to idx 14 ? even if it means a low score for pixels of OOD object ?