我已经在Keras中为55个二进制类别训练了一个模型。该模型无法拟合训练数据。在训练过程中，大多数类别的准确性都提高了（损失减少了），但是对于某些类别，训练的准确性没有超过％55和％60。
据我所知，至少所有Train_acc都应接近％100（但不一定Best_val_acc），因此我可以说该网络是经过训练和过拟合的（我知道我们需要避免过度拟合）。但是，这种结果表明网络无法学习我的训练数据。编码时我做错什么了吗？我在这篇文章的结尾附上我的网络摘要。
我把部分结果供您审核。顺便说一句，我看了一下数据集的注释，发现与Age31-45和Age17-30相比，AgeLess16和Age46-60的样本最多。
Accuracy and Loss for Femal_FineTuned_AllLayers:
Train_acc: %96.35855110561212, Train_loss: 0.10732883015214066
Best_val_acc: %93.82214298728499, Val_loss: 0.174280562132292
**********
**********
Accuracy and Loss for AgeLess16_FineTuned_AllLayers:
Train_acc: %99.17721500686592, Train_loss: 0.05093206035518569
Best_val_acc: %99.2638396430493, Val_loss: 0.04364626104115246
**********
**********
Accuracy and Loss for Age17-30_FineTuned_AllLayers:
Train_acc: %59.80718333591055, Train_loss: 0.6711170782595829
Best_val_acc: %60.8303891317162, Val_loss: 0.6664492566186773
**********
**********
Accuracy and Loss for Age31-45_FineTuned_AllLayers:
Train_acc: %55.24286194234288, Train_loss: 0.6829969891255696
Best_val_acc: %56.93757418889741, Val_loss: 0.6803399343723684
**********
**********
Accuracy and Loss for Age46-60_FineTuned_AllLayers:
Train_acc: %96.42478600210185, Train_loss: 0.1544494673218781
Best_val_acc: %96.95524095576279, Val_loss: 0.13411041373305804
**********
**********
Accuracy and Loss for BodyFat_FineTuned_AllLayers:
Train_acc: %86.43950484528554, Train_loss: 0.39786816761884614
Best_val_acc: %87.1849227551858, Val_loss: 0.3838976073145726
**********
**********
Accuracy and Loss for BodyNormal_FineTuned_AllLayers:
Train_acc: %77.93935805949562, Train_loss: 0.5303131385909517
Best_val_acc: %77.9151941672652, Val_loss: 0.528312525914892
**********
**********
Accuracy and Loss for BodyThin_FineTuned_AllLayers:
Train_acc: %92.5551949922639, Train_loss: 0.2664116882610258
Best_val_acc: %92.64428666118178, Val_loss: 0.2637324569959663
**********
**********
Accuracy and Loss for Customer_FineTuned_AllLayers:
Train_acc: %95.77274004045148, Train_loss: 0.15548874558041995
Best_val_acc: %96.18963424392808, Val_loss: 0.14634984320751418
**********
**********
Accuracy and Loss for Employee_FineTuned_AllLayers:
Train_acc: %96.50132414397682, Train_loss: 0.1296264658043803
Best_val_acc: %96.8138981448186, Val_loss: 0.12327752215623224
**********
**********
Accuracy and Loss for hs-BaldHead_FineTuned_AllLayers:
Train_acc: %99.40241376529696, Train_loss: 0.03519799720703804
Best_val_acc: %99.5229680876569, Val_loss: 0.02903259372079769
**********
**********
Accuracy and Loss for hs-LongHair_FineTuned_AllLayers:
Train_acc: %90.63732630326743, Train_loss: 0.2004715618580678
Best_val_acc: %90.86572360795856, Val_loss: 0.20626683917924438
**********
**********
Accuracy and Loss for hs-BlackHair_FineTuned_AllLayers:
Train_acc: %93.31027304104997, Train_loss: 0.22499078004035636
Best_val_acc: %93.75147156426145, Val_loss: 0.22317562431965826
**********
**********
Accuracy and Loss for ub-Shirt_FineTuned_AllLayers:
Train_acc: %78.37650845195218, Train_loss: 0.5004801768453149
Best_val_acc: %79.33451089441988, Val_loss: 0.48233152837333465
**********
**********
Accuracy and Loss for ub-Sweater_FineTuned_AllLayers:
Train_acc: %92.05180968884198, Train_loss: 0.2790391942859573
Best_val_acc: %92.72673661843626, Val_loss: 0.25994354528588876
**********
**********
Accuracy and Loss for ub-Vest_FineTuned_AllLayers:
Train_acc: %95.41065580517817, Train_loss: 0.17248446327815925
Best_val_acc: %95.83038808459808, Val_loss: 0.16092383824025575
**********
**********
Accuracy and Loss for ub-TShirt_FineTuned_AllLayers:
Train_acc: %76.87224004278384, Train_loss: 0.5403878908250168
Best_val_acc: %77.37338032893774, Val_loss: 0.5337123219691682
**********
**********
Accuracy and Loss for ub-Cotton_FineTuned_AllLayers:
Train_acc: %88.4795401174074, Train_loss: 0.35894714934109306
Best_val_acc: %89.44640678778133, Val_loss: 0.3402983718524271
**********
**********
Accuracy and Loss for ub-Jacket_FineTuned_AllLayers:
Train_acc: %71.56020030159968, Train_loss: 0.5912178378658572
Best_val_acc: %71.77856311342461, Val_loss: 0.5866910439638423
**********
**********
Accuracy and Loss for ub-SuitUp_FineTuned_AllLayers:
Train_acc: %97.03414728176884, Train_loss: 0.13934014985967874
Best_val_acc: %97.10836230093796, Val_loss: 0.1311324757366969
**********
**********
部分代码：
input_shape= (500,500, 3)
learning_rate=0.01
Number_of_epochs_all_layers = 20
DROPOUT_PROB = 0.5
BATCH_SIZE = 10
output_layers[i] = KL.Dense(1, activation='sigmoid', name=Categories[i])(x)
model.compile(optimizer=optimizers.SGD(lr=learning_rate),
loss='binary_crossentropy',
loss_weights = [0.5, 0.25, 0.25, 0.25, 0.25, 0.333, 0.333, 0.333, 0.5, 0.5,
0.333, 0.333, 0.333, 0.5, 0.5, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.166, 0.166, 0.166, 0.166, 0.166, 0.166, 0.166, 0.166, 0.166, 0.166, 0.166, 0.166,
0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125, 0.125,
0.111, 0.111, 0.111, 0.111, 0.111, 0.111, 0.111, 0.111, 0.111, 0.5],
metrics=['accuracy'])
模型的一部分：
_________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) (None, 500, 500, 3) 0
__________________________________________________________________________________________________
bn4f_branch2b (BatchNorm) (None, 32, 32, 256) 1024 res4f_branch2b[0][0]
__________________________________________________________________________________________________
activation_27 (Activation) (None, 32, 32, 256) 0 bn4f_branch2b[0][0]
__________________________________________________________________________________________________
res4f_branch2c (Conv2D) (None, 32, 32, 1024) 263168 activation_27[0][0]
__________________________________________________________________________________________________
bn4f_branch2c (BatchNorm) (None, 32, 32, 1024) 4096 res4f_branch2c[0][0]
__________________________________________________________________________________________________
input_2 (InputLayer) (None, 32, 32, 3) 0
__________________________________________________________________________________________________
add_13 (Add) (None, 32, 32, 1024) 0 bn4f_branch2c[0][0]
res4e_out[0][0]
__________________________________________________________________________________________________
dense_1 (Dense) (None, 32, 32, 1024) 4096 input_2[0][0]
__________________________________________________________________________________________________
res4f_out (Activation) (None, 32, 32, 1024) 0 add_13[0][0]
__________________________________________________________________________________________________
dropout_1 (Dropout) (None, 32, 32, 1024) 0 dense_1[0][0]
__________________________________________________________________________________________________
multiply_1 (Multiply) (None, 32, 32, 1024) 0 res4f_out[0][0]
dropout_1[0][0]
__________________________________________________________________________________________________
global_average_pooling2d_1 (Glo (None, 1024) 0 multiply_1[0][0]
__________________________________________________________________________________________________
dense_2 (Dense) (None, 256) 262400 global_average_pooling2d_1[0][0]
__________________________________________________________________________________________________
dropout_2 (Dropout) (None, 256) 0 dense_2[0][0]
__________________________________________________________________________________________________
dense_3 (Dense) (None, 256) 65792 dropout_2[0][0]
__________________________________________________________________________________________________
dropout_3 (Dropout) (None, 256) 0 dense_3[0][0]
__________________________________________________________________________________________________
dense_4 (Dense) (None, 128) 32896 dropout_3[0][0]
__________________________________________________________________________________________________
dense_12 (Dense) (None, 128) 32896 dropout_3[0][0]
__________________________________________________________________________________________________
dense_6 (Dense) (None, 128) 32896 dropout_3[0][0]
__________________________________________________________________________________________________
dense_8 (Dense) (None, 128) 32896 dropout_3[0][0]
__________________________________________________________________________________________________
dense_10 (Dense) (None, 128) 32896 dropout_3[0][0]
__________________________________________________________________________________________________
dense_14 (Dense) (None, 128) 32896 dropout_3[0][0]
__________________________________________________________________________________________________
dense_16 (Dense) (None, 128) 32896 dropout_3[0][0]
__________________________________________________________________________________________________
dropout_4 (Dropout) (None, 128) 0 dense_4[0][0]
__________________________________________________________________________________________________
dropout_12 (Dropout) (None, 128) 0 dense_12[0][0]
__________________________________________________________________________________________________
dropout_6 (Dropout) (None, 128) 0 dense_6[0][0]
__________________________________________________________________________________________________
dropout_8 (Dropout) (None, 128) 0 dense_8[0][0]
__________________________________________________________________________________________________
dropout_10 (Dropout) (None, 128) 0 dense_10[0][0]
__________________________________________________________________________________________________
dropout_14 (Dropout) (None, 128) 0 dense_14[0][0]
__________________________________________________________________________________________________
dropout_16 (Dropout) (None, 128) 0 dense_16[0][0]
__________________________________________________________________________________________________
dense_5 (Dense) (None, 64) 8256 dropout_4[0][0]
__________________________________________________________________________________________________
dense_13 (Dense) (None, 64) 8256 dropout_12[0][0]
__________________________________________________________________________________________________
dense_7 (Dense) (None, 64) 8256 dropout_6[0][0]
__________________________________________________________________________________________________
dense_9 (Dense) (None, 64) 8256 dropout_8[0][0]
__________________________________________________________________________________________________
dense_11 (Dense) (None, 64) 8256 dropout_10[0][0]
__________________________________________________________________________________________________
dense_15 (Dense) (None, 64) 8256 dropout_14[0][0]
__________________________________________________________________________________________________
dense_17 (Dense) (None, 64) 8256 dropout_16[0][0]
__________________________________________________________________________________________________
dropout_5 (Dropout) (None, 64) 0 dense_5[0][0]
__________________________________________________________________________________________________
dropout_13 (Dropout) (None, 64) 0 dense_13[0][0]
__________________________________________________________________________________________________
dropout_7 (Dropout) (None, 64) 0 dense_7[0][0]
__________________________________________________________________________________________________
dropout_9 (Dropout) (None, 64) 0 dense_9[0][0]
__________________________________________________________________________________________________
dropout_11 (Dropout) (None, 64) 0 dense_11[0][0]
__________________________________________________________________________________________________
dropout_15 (Dropout) (None, 64) 0 dense_15[0][0]
__________________________________________________________________________________________________
dropout_17 (Dropout) (None, 64) 0 dense_17[0][0]
__________________________________________________________________________________________________
Femal (Dense) (None, 1) 65 dropout_5[0][0]
__________________________________________________________________________________________________
AgeLess16 (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
Age17-30 (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
Age31-45 (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
Age46-60 (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
BodyFat (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
BodyNormal (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
BodyThin (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
Customer (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
Employee (Dense) (None, 1) 65 dropout_13[0][0]
__________________________________________________________________________________________________
hs-BaldHead (Dense) (None, 1) 65 dropout_5[0][0]
__________________________________________________________________________________________________
hs-LongHair (Dense) (None, 1) 65 dropout_5[0][0]
__________________________________________________________________________________________________
hs-BlackHair (Dense) (None, 1) 65 dropout_5[0][0]
__________________________________________________________________________________________________
hs-Hat (Dense) (None, 1) 65 dropout_5[0][0]
__________________________________________________________________________________________________
hs-Glasses (Dense) (None, 1) 65 dropout_5[0][0]
__________________________________________________________________________________________________
ub-Shirt (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-Sweater (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-Vest (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-TShirt (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-Cotton (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-Jacket (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-SuitUp (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-Tight (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-ShortSleeve (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
ub-Others (Dense) (None, 1) 65 dropout_7[0][0]
__________________________________________________________________________________________________
lb-LongTrousers (Dense) (None, 1) 65 dropout_9[0][0]
__________________________________________________________________________________________________
lb-Skirt (Dense) (None, 1) 65 dropout_9[0][0]
__________________________________________________________________________________________________
lb-ShortSkirt (Dense) (None, 1) 65 dropout_9[0][0]
__________________________________________________________________________________________________
lb-Dress (Dense) (None, 1) 65 dropout_9[0][0]
__________________________________________________________________________________________________
lb-Jeans (Dense) (None, 1) 65 dropout_9[0][0]
__________________________________________________________________________________________________
lb-TightTrousers (Dense) (None, 1) 65 dropout_9[0][0]
__________________________________________________________________________________________________
shoes-Leather (Dense) (None, 1) 65 dropout_11[0][0]
__________________________________________________________________________________________________
shoes-Sports (Dense) (None, 1) 65 dropout_11[0][0]
__________________________________________________________________________________________________
shoes-Boots (Dense) (None, 1) 65 dropout_11[0][0]
__________________________________________________________________________________________________
shoes-Cloth (Dense) (None, 1) 65 dropout_11[0][0]
__________________________________________________________________________________________________
shoes-Casual (Dense) (None, 1) 65 dropout_11[0][0]
__________________________________________________________________________________________________
shoes-Other (Dense) (None, 1) 65 dropout_11[0][0]
__________________________________________________________________________________________________
attachment-Backpack (Dense) (None, 1) 65 dropout_15[0][0]
__________________________________________________________________________________________________
attachment-ShoulderBag (Dense) (None, 1) 65 dropout_15[0][0]
__________________________________________________________________________________________________
attachment-HandBag (Dense) (None, 1) 65 dropout_15[0][0]
__________________________________________________________________________________________________
attachment-Box (Dense) (None, 1) 65 dropout_15[0][0]
__________________________________________________________________________________________________
attachment-PlasticBag (Dense) (None, 1) 65 dropout_15[0][0]
__________________________________________________________________________________________________
attachment-PaperBag (Dense) (None, 1) 65 dropout_15[0][0]
__________________________________________________________________________________________________
attachment-HandTrunk (Dense) (None, 1) 65 dropout_15[0][0]
__________________________________________________________________________________________________
attachment-Other (Dense) (None, 1) 65 dropout_15[0][0]
__________________________________________________________________________________________________
action-Calling (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
action-Talking (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
action-Gathering (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
action-Holding (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
action-Pushing (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
action-Pulling (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
action-CarryingByArm (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
action-CarryingByHand (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
action-Other (Dense) (None, 1) 65 dropout_17[0][0]
__________________________________________________________________________________________________
Male (Dense) (None, 1) 65 dropout_5[0][0]