Epoch [20/20], Iter [1055/1057] Loss: 1.8820
loc_loss:0.984644 conf_loss:1.205544, pos_num:78 loc_loss:0.698231 conf_loss:1.439163, pos_num:163 loc_loss:0.644837 conf_loss:1.328891, pos_num:143 loc_loss:0.420935 conf_loss:1.332006, pos_num:149 loc_loss:0.246193 conf_loss:0.883853, pos_num:242 Epoch [20/20], Iter [1055/1057] Loss: 1.1300
loc_loss:1.417856 conf_loss:2.061183, pos_num:152 loc_loss:0.655475 conf_loss:1.698712, pos_num:65 loc_loss:1.235528 conf_loss:2.012040, pos_num:144 loc_loss:0.798853 conf_loss:1.713999, pos_num:109 loc_loss:0.823239 conf_loss:1.953249, pos_num:318 Epoch [50/50], Iter [1700/1701] Loss: 2.7765, average_loss: 2.8849
loc_loss:0.802718 conf_loss:1.943955, pos_num:284 loc_loss:0.867129 conf_loss:1.820582, pos_num:420 loc_loss:0.885825 conf_loss:1.830107, pos_num:358 loc_loss:0.811850 conf_loss:1.881572, pos_num:501 loc_loss:0.975667 conf_loss:1.921641, pos_num:540 Epoch [50/50], Iter [680/681] Loss: 2.8973, average_loss: 2.5820
46 2.27359845486 47 2.27207741518 48 2.26043195595 49 2.26634732234
loc_loss:2.556887 conf_loss:2.489368, pos_num:68 loc_loss:2.234761 conf_loss:2.448641, pos_num:111 loc_loss:2.569000 conf_loss:2.495923, pos_num:105 loc_loss:2.542970 conf_loss:2.470961, pos_num:74 loc_loss:2.530408 conf_loss:2.485945, pos_num:79 Epoch [1/50], Iter [1045/1677] Loss: 5.0164, average_loss: 5.0478 loc_loss:2.033609 conf_loss:2.263520, pos_num:173 loc_loss:2.454619 conf_loss:2.263516, pos_num:109 loc_loss:2.247968 conf_loss:2.262204, pos_num:257 loc_loss:2.366087 conf_loss:2.263796, pos_num:50 loc_loss:2.315195 conf_loss:2.261710, pos_num:192 Epoch [10/50], Iter [735/1677] Loss: 4.5769, average_loss: 4.5744
- 修改1
复现时，2倍致密是原scale的1/4，是2倍scale的anchor的1/8，写成了1/4，已更改 - 修改2
为每一个box label都添加了与之IOU最大的box，不管IOU阈值是多少，这样导致有inf loc loss出现，是因为targets的宽和高有的为0，也就是dataset代码中random_crop有问题，添加了对box_label的宽和高限制为10像素后，问题不再出现。 - 修改3 使用Adam
不明白为什么突然loss爆炸了
loc_loss:115.657501 conf_loss:39.798553, pos_num:2528
<!-- 300 epoch -->
Epoch [300/300], Iter [400/403] Loss: 3.4930, average_loss: 3.5764
loc_loss:1.732548 conf_loss:1.807370, pos_num:1120
loc_loss:1.832072 conf_loss:2.002608, pos_num:1711
loc_loss:1.265184 conf_loss:1.525407, pos_num:624