search

vita-epfl / bounding-box-prediction

Bounding box prediction library. Official implementation of two papers on human 2D/3D bounding box prediction
48 stars 14 forks source link

imbalanced cross/ not cross labels #6

Closed jinxing94 closed 2 years ago

jinxing94 commented 2 years ago

I do not know why I find that only about 10% samples are with cross labels. Could you help check the ratio of positive samples ?

If the data is indeed imbalanced, other metrics like the recall, f1-score of intention prediction may be important.

SaeedSaadatnejad commented 2 years ago

Hello,

I don't know the exact ratio but it is true that there are fewer crossing pedestrians in the dataset. Indeed, what you suggested (reporting recall and f1-score) is a very good idea for a comprehensive study. We only reported the accuracy and compared it with previous works.

Best, Saeed