Infer higher FPs?

Paper: [https://gangfan.github.io/assets/papers/gang_smoke_icse2019_preprint.pdf](url)
Hey, some points are:
1. A manual check reveals that INFER generates a specification for each analyzed function, and discards some of these specifications for unknown reasons.
2. On average, SMOKE achieves the highest efficiency and is significantly (>5.2X faster than SABER, >13.0X faster than PINPOINT, >12.4X faster than CSA, >22.8X faster than INFER) faster than other tools. SMOKE performs better than SABER and PINPOINT in all subjects.
3. On average, SMOKE achieves an FP rate of 24.4%, which is the lowest one among all the tools. As discussed in Section V-B, not all the subjects can be successfully analyzed by all the tools. And the number of false positives in the failure cases (e.g., SABER fails in the project LAME) is counted as 0. For the medium-sized subject, the small number of failure cases (only 2 by SABER, 1 by CSA and INFER ) allows a fair comparison of all the tools, where SMOKE outperforms all other tools with a lowest FP rate of 15.0%. The second best tool, PINPOINT, achieves a much higher FP rate of 41.7%. INFER performs the worst, with an FP rate of 87.5%.

Here, Infer team does any benchmark test about memory leak or other checkers?