Hello, I want to train a model, which uses an image as an input and predicts a float number between 0 and 360. When training the model, training loss and validation loss are not decreasing.
I used the following Model:
torch.autograd.set_detect_anomaly(True)
import torch.nn as nn
import torch.nn.functional as F
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(3, 16, 3, padding=1)
self.conv2 = nn.Conv2d(16, 32, 3, padding=1)
self.conv3 = nn.Conv2d(32, 64, 3, padding=1)
self.pool = nn.MaxPool2d(2, 2)
self.fc1 = nn.Linear(64 * 16 * 16, 256)
self.fc2 = nn.Linear(256, 1)
self.dropout = nn.Dropout(0.25)
def forward(self, x):
# add sequence of convolutional and max pooling layers
x = self.pool(F.relu(self.conv1(x)))
x = self.pool(F.relu(self.conv2(x)))
x = self.pool(F.relu(self.conv3(x)))
x = x.view(-1,64 * 16 * 16)
x = self.dropout(x)
x = F.relu(self.fc1(x))
#x = self.fc1(x)
x = self.dropout(x)
x = self.fc2(x)
return x
if train_on_gpu:
model.cuda()
The images are normalized and resized to 128 x 128. I tried approaches that I found in this forum, including learning rate scheduling, but that did not work for me.
import torch.optim as optim
from torch.optim.lr_scheduler import StepLR,ReduceLROnPlateau
optimizer = optim.SGD(model.parameters(),lr=0.001,momentum=0.9,weight_decay=0.01, nesterov=True)
scheduler=ReduceLROnPlateau(optimizer, mode='max', factor=0.7, patience=3,verbose=True)
criterion = nn.MSELoss().cuda()
I’m training with a batch size of 32. My train and eval code looks like this:
n_epochs = 50
valid_loss_min = np.Inf
for epoch in range(1, n_epochs+1):
train_loss = 0.0
valid_loss = 0.0
model.train()
for data, target in tqdm(train_loader):
if train_on_gpu:
data, target = data.cuda(), target.cuda()
optimizer.zero_grad()
#print(label.data)
output = model(data)
target = target.unsqueeze(1)
loss = criterion(output.float(), target.float())
loss.backward()
optimizer.step()
train_loss += loss.item()*data.size(0)
model.eval()
for data, target in valid_loader:
if train_on_gpu:
data, target = data.cuda(), target.cuda()
output = model(data)
target = target.unsqueeze(1)
loss = criterion(output.float(), target.float())
valid_loss += loss.item()*data.size(0)
train_loss = train_loss/len(train_loader.sampler)
valid_loss = valid_loss/len(valid_loader.sampler)
print('Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f}'.format(
epoch, train_loss, valid_loss))
if valid_loss <= valid_loss_min:
print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(
valid_loss_min,
valid_loss))
torch.save(model.state_dict(), 'model.pt')
valid_loss_min = valid_loss
scheduler.step(valid_loss)
I could not find the mistake in my code. Is my model not suited for image regression tasks? If so, what would I have to modify? Until now I only worked with classifiers.
Thank you