runtime question

[zbwustc]

i trained faster rcnn on my own vehicle dataset,and used demo to test.But i found it strange about runtime.
faster_rcnn-master: test (voc_2007_test) 3994/4022 time: 0.058s
faster_rcnn-master: test (voc_2007_test) 3995/4022 time: 0.106s
faster_rcnn-master: test (voc_2007_test) 3996/4022 time: 0.057s
faster_rcnn-master: test (voc_2007_test) 3997/4022 time: 0.059s
faster_rcnn-master: test (voc_2007_test) 3998/4022 time: 0.109s
faster_rcnn-master: test (voc_2007_test) 3999/4022 time: 0.112s
faster_rcnn-master: test (voc_2007_test) 4000/4022 time: 0.060s
datasets\VOCdevkit2007\VOC2007\JPEGImages\008041.jpg (1600x1200): time 0.028s (resize+conv+proposal: 0.016s, nms+regionwise: 0.012s) datasets\VOCdevkit2007\VOC2007\JPEGImages\008042.jpg (1600x1200): time 0.028s (resize+conv+proposal: 0.017s, nms+regionwise: 0.011s) datasets\VOCdevkit2007\VOC2007\JPEGImages\008043.jpg (1600x1200): time 0.028s (resize+conv+proposal: 0.017s, nms+regionwise: 0.011s) datasets\VOCdevkit2007\VOC2007\JPEGImages\008044.jpg (1600x1200): time 0.029s (resize+conv+proposal: 0.017s, nms+regionwise: 0.012s) mean time: 0.029s Cleared 0 solvers and 2 stand-alone nets
the first is in the trian log,it is the result of test dataset when trainning.the second is the results when using demo.m.
why is the time so different?demo uses less time a lot.