Caffe-ssd訓練MobileNet模型過程中列印“ Missing true_pos for label:”的提示

模型訓練過程中列印出的資訊如下：

I1118 12:32:38.050251 179535 solver.cpp:259]     Train net output #0: mbox_loss = 1.16632 (* 1 = 1.16632 loss)
I1118 12:32:38.413388 179535 sgd_solver.cpp:138] Iteration 42990, lr = 0.00499549
I1118 12:33:22.066315 179535 solver.cpp:603] Snapshotting to binary proto file snapshot/mobilenet_iter_43000.caffemodel
I1118 12:33:22.158694 179535 sgd_solver.cpp:307] Snapshotting solver state to binary proto file snapshot/mobilenet_iter_43000.solverstate
I1118 12:33:22.182749 179535 solver.cpp:433] Iteration 43000, Testing net (#0)
I1118 12:33:22.183234 179535 net.cpp:693] Ignoring source layer mbox_loss
W1118 12:33:46.411466 179535 solver.cpp:524] Missing true_pos for label: 6
W1118 12:33:46.411643 179535 solver.cpp:524] Missing true_pos for label: 14
W1118 12:33:46.411772 179535 solver.cpp:524] Missing true_pos for label: 20
I1118 12:33:46.411809 179535 solver.cpp:553]     Test net output #0: detection_eval = 0.634302
I1118 12:33:50.749930 179535 solver.cpp:243] Iteration 43000, loss = 1.74211
I1118 12:33:50.750015 179535 solver.cpp:259]     Train net output #0: mbox_loss = 2.42199 (* 1 = 2.42199 loss)
           

為什麼會提示這條資訊呢？

查了一下資料

原來是對應的那幾類的訓練圖像的數量太少了，以至于不能有效的被檢測出，可以忽略

感謝大神的解答

參考

【1】When train with coco dataset, print warning info: “Missing true_pos for label: 71”. #717