Omitted Labels

red high­light­ed objects/persons were miss­ing in a dataset cru­cial for autonomous dri­ving

Brad Dwyer found a lot of miss­ing or omit­ted labels in a set that is used for train­ing and test­ing autonomous dryv­ing sys­tems. »We did a hand-check of the 15,000 images in the wide­ly used Udac­i­ty Dataset 2 and found prob­lems with 4,986 (33%) of them.« Since this is a Open Source Dataset used pri­mar­i­ly for edu­ca­tion­al pur­pos­es, but as the author found out obvi­ous­ly also for test cars on pub­lic streets, he pub­lished a cor­rect­ed set at https://public.roboflow.ai/object-detection/self-driving-car

Besides the hon­or­able work of Dwyer, these omis­sions lead to the larg­er ques­tion of the reli­abli­ty of many data sets which are being used for train­ing.