I think the problem is not that the car is cut off, its that the cut-off portion of the car isn't labeled. If that's your training data, then you're training t

I think the problem is not that the car is cut off, its that the cut-off portion of the car isn't labeled. If that's your training data, then you're training the system to interpret cut-off bits of cars as not cars.