The loss function converges slowly and the test accuracy is not high??

[lunalulu]

When I trained this network with the person data set(trian data number 2785,test data number 310) I collected myself, the loss function converges slowly and the test accuracy is not high. In addition, the person figure in the data set has more occlusion. When iterating 75, 000 times, loss is still between 1.5 and 1.8, and mbox_loss is between 1.4 and 2.2. test detection_eval like this, Test each iteration 5,000 times

I1030 14:04:02.521121 14101 solver.cpp:546] Test net output #0: detection_eval = 0.799385 I1030 18:18:31.173002 14101 solver.cpp:546] Test net output #0: detection_eval = 0.830706 I1030 22:29:47.758836 14101 solver.cpp:546] Test net output #0: detection_eval = 0.842571 I1031 02:41:34.874645 14101 solver.cpp:546] Test net output #0: detection_eval = 0.854919 I1031 06:54:12.981827 14101 solver.cpp:546] Test net output #0: detection_eval = 0.860794 I1031 11:05:26.106914 14101 solver.cpp:546] Test net output #0: detection_eval = 0.868059 I1031 15:17:35.631048 14101 solver.cpp:546] Test net output #0: detection_eval = 0.860603 I1031 19:28:55.489873 14101 solver.cpp:546] Test net output #0: detection_eval = 0.863569 I1031 23:40:01.686856 14101 solver.cpp:546] Test net output #0: detection_eval = 0.864795 I1101 03:52:07.592077 14101 solver.cpp:546] Test net output #0: detection_eval = 0.865677 I1101 08:02:45.661051 14101 solver.cpp:546] Test net output #0: detection_eval = 0.864265 I1101 12:14:21.095113 14101 solver.cpp:546] Test net output #0: detection_eval = 0.868086 I1101 16:26:37.139066 14101 solver.cpp:546] Test net output #0: detection_eval = 0.866303 I1101 20:38:43.176023 14101 solver.cpp:546] Test net output #0: detection_eval = 0.868358 I1102 00:50:53.074564 14101 solver.cpp:546] Test net output #0: detection_eval = 0.867538

thanks very much!