Even after being edited, it's still wrong. It shows the significantly lower Inception4 performance as a "40% speedup" instead of 40% of baseline images/sec.