For the non-object-centered datasets CompCars and Churches, we follow [18] and add random translation offsets during training.
確定! 回上一頁