The training occurs NAN and -inf, anyone knows what happened here?Iteration 20, loss = -nan I0905 19:18:20.666388 10119 solver.cpp:259] Train net output #0: mbox_loss = inf (* 1 = inf loss) I0905 19:18:20.666396 10119 sgd_solver.cpp:138] Iteration 20, lr = 0.05

[xjiao004]