Figure 5: Some typical photos that lead to false-negative input (participants chose “Diff” when the pairs actually are the same).