Hello everyone, When I run my model, it gives me the following result, my accuracy in training increases, and the error in training decreases, but when I validate, it does not increase or decrease. Can someone guide me?
Epoch 0/19
----------
train Loss: 0.2080 Acc: 0.9127
val Loss: 1.9388 Acc: 0.5716
Epoch 1/19
----------
train Loss: 0.2081 Acc: 0.9117
val Loss: 1.9498 Acc: 0.5949
Epoch 2/19
----------
train Loss: 0.2057 Acc: 0.9125
val Loss: 2.0436 Acc: 0.5814
Epoch 3/19
----------
train Loss: 0.2102 Acc: 0.9111
val Loss: 1.6863 Acc: 0.5973
Epoch 4/19
----------
train Loss: 0.2064 Acc: 0.9108
val Loss: 1.8532 Acc: 0.5802
Epoch 5/19
----------
train Loss: 0.2050 Acc: 0.9112
val Loss: 1.9416 Acc: 0.5667
Epoch 6/19
----------
train Loss: 0.1812 Acc: 0.9212
val Loss: 1.8484 Acc: 0.5814
Epoch 7/19
----------
train Loss: 0.1644 Acc: 0.9293
val Loss: 1.8913 Acc: 0.5802
Epoch 8/19
----------
train Loss: 0.1548 Acc: 0.9340
val Loss: 1.8807 Acc: 0.5716
Epoch 9/19
----------
train Loss: 0.1585 Acc: 0.9328
val Loss: 1.8908 Acc: 0.5728
Epoch 10/19
----------
train Loss: 0.1533 Acc: 0.9352
val Loss: 1.9874 Acc: 0.5814
Epoch 11/19
----------
train Loss: 0.1450 Acc: 0.9399
val Loss: 2.0618 Acc: 0.5949
Epoch 12/19
----------
train Loss: 0.1464 Acc: 0.9406
val Loss: 2.1211 Acc: 0.5789
Epoch 13/19
----------
train Loss: 0.1447 Acc: 0.9364
val Loss: 2.1088 Acc: 0.5802
Epoch 14/19
----------
train Loss: 0.1421 Acc: 0.9386
val Loss: 2.0540 Acc: 0.5851
Epoch 15/19
----------
train Loss: 0.1483 Acc: 0.9376
val Loss: 2.1731 Acc: 0.5851
Epoch 16/19
----------
train Loss: 0.1420 Acc: 0.9408
val Loss: 2.1140 Acc: 0.5741
Epoch 17/19
----------
train Loss: 0.1516 Acc: 0.9346
val Loss: 2.1321 Acc: 0.5741
model = models.resnet152(pretrained=True)
batch_size=32
optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.001, momentum=0.9)
exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)
The problem is about binary classification:
ADHD = Attention-deficit/hyperactivity disorder.
Control = Non-ADHD.
train > ADHD > 3800 Images
>Control>4100 Images
val > ADHD > 608 Images
>Control > 692 Images
Is the number of validation images large enough?