The quantitative prediction of nonadiabatic transitions between different electronic states is important to understand ultrafast processes in photochemistry. A variety of mixed quantum-classical molecular dynamics methods such as surface hopping and Ehrenfest mean-field have been developed. However, how to choose an appropriate one from a wide diversity of dynamics algorithms to study a realistic photochemical process is still unclear. In this work, we implemented 30 combinations of different mixed quantum-classical dynamics methods, including 24 surface hopping models with 8 decoherence corrections and 3 momentum rescaling strategies as well as 6 mean-field models. Then we performed numerical investigations by simulating the photoisomerization of cis-azobenzene combined with on-the-fly electronic structure calculations. Predictions of the S1 lifetime and the quantum yield of the photoproduct using different models are distinct. Surface hopping is more robust than mean-field in our test system. Moreover, the choice of momentum rescaling methods in surface hopping brings more significant changes than decoherence corrections, while a large discrepancy between simulation results with different mean-field algorithms has been observed.