FPN訓練自己的小目標數據爬坑日誌（2）

錯誤：RcnnLossBBox = 0  

I0625 20:40:38.710259 18430 sgd_solver.cpp:107] Iteration 2640, lr = 0.001

I0625 20:40:50.785377 18430 solver.cpp:229] Iteration 2660, loss = 0.262648
I0625 20:40:50.785401 18430 solver.cpp:245]     Train net output #0: FPNClsLoss = 0.0808626 (* 1 = 0.0808626 loss)
I0625 20:40:50.785408 18430 solver.cpp:245]     Train net output #1: FPNLossBBox = 0.277722 (* 1 = 0.277722 loss)
I0625 20:40:50.785411 18430 solver.cpp:245]     Train net output #2: RcnnLossBBox = 0 (* 1 = 0 loss)

I0625 20:40:50.785415 18430 solver.cpp:245]     Train net output #3: RcnnLossCls = 1.07288e-06 (* 1 = 1.07288e-06 loss)

 

 

loss_bbox = 0，解決後續補充

1、嘗試過將base_lr學習率修改爲0.00001

2、因爲本人得顯卡是k2200故將batch_size設置爲1，最後換了1080ti，loss不爲0（基本解決）

3、後續嘗試將數據集擴大