Spaces:
Running
Running
Update pages/19_ResNet.py
Browse files- pages/19_ResNet.py +104 -16
pages/19_ResNet.py
CHANGED
|
@@ -35,25 +35,18 @@ transform = transforms.Compose([
|
|
| 35 |
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
|
| 36 |
])
|
| 37 |
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
val_size = len(subset_indices) - train_size
|
| 45 |
-
|
| 46 |
-
train_indices = subset_indices[:train_size]
|
| 47 |
-
val_indices = subset_indices[train_size:]
|
| 48 |
-
|
| 49 |
-
train_dataset = Subset(train_dataset, train_indices)
|
| 50 |
-
val_dataset = Subset(val_dataset, val_indices)
|
| 51 |
|
| 52 |
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=4)
|
| 53 |
val_loader = DataLoader(val_dataset, batch_size=batch_size, shuffle=False, num_workers=4)
|
| 54 |
|
| 55 |
dataloaders = {'train': train_loader, 'val': val_loader}
|
| 56 |
-
class_names =
|
| 57 |
|
| 58 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
| 59 |
|
|
@@ -78,8 +71,103 @@ imshow(out, title=[class_names[x] for x in classes])
|
|
| 78 |
# Model Preparation Section
|
| 79 |
st.markdown("""
|
| 80 |
### Model Preparation
|
| 81 |
-
We will use a pre-trained ResNet-18 model and fine-tune the final fully connected layer to match the number of classes in our dataset.
|
| 82 |
""")
|
| 83 |
|
| 84 |
# Load Pre-trained ResNet Model
|
| 85 |
-
model_ft = models.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
|
| 36 |
])
|
| 37 |
|
| 38 |
+
full_dataset = datasets.CIFAR10(root='./data', train=True, download=True, transform=transform)
|
| 39 |
+
subset_indices = list(range(1000)) # Use only 1000 samples for simplicity
|
| 40 |
+
subset_dataset = Subset(full_dataset, subset_indices)
|
| 41 |
+
train_size = int(0.8 * len(subset_dataset))
|
| 42 |
+
val_size = len(subset_dataset) - train_size
|
| 43 |
+
train_dataset, val_dataset = torch.utils.data.random_split(subset_dataset, [train_size, val_size])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
|
| 45 |
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=4)
|
| 46 |
val_loader = DataLoader(val_dataset, batch_size=batch_size, shuffle=False, num_workers=4)
|
| 47 |
|
| 48 |
dataloaders = {'train': train_loader, 'val': val_loader}
|
| 49 |
+
class_names = full_dataset.classes
|
| 50 |
|
| 51 |
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
| 52 |
|
|
|
|
| 71 |
# Model Preparation Section
|
| 72 |
st.markdown("""
|
| 73 |
### Model Preparation
|
| 74 |
+
We will use a pre-trained ResNet-18 model and fine-tune the final fully connected layer to match the number of classes in our custom dataset.
|
| 75 |
""")
|
| 76 |
|
| 77 |
# Load Pre-trained ResNet Model
|
| 78 |
+
model_ft = models.resnet18(pretrained=True)
|
| 79 |
+
num_ftrs = model_ft.fc.in_features
|
| 80 |
+
model_ft.fc = nn.Linear(num_ftrs, len(class_names))
|
| 81 |
+
|
| 82 |
+
model_ft = model_ft.to(device)
|
| 83 |
+
|
| 84 |
+
# Define Loss Function and Optimizer
|
| 85 |
+
criterion = nn.CrossEntropyLoss()
|
| 86 |
+
optimizer_ft = optim.SGD(model_ft.parameters(), lr=learning_rate, momentum=0.9)
|
| 87 |
+
|
| 88 |
+
# Training Section
|
| 89 |
+
st.markdown("""
|
| 90 |
+
### Training
|
| 91 |
+
We will train the model using stochastic gradient descent (SGD) with a learning rate scheduler. The training and validation loss and accuracy will be plotted to monitor the training process.
|
| 92 |
+
""")
|
| 93 |
+
|
| 94 |
+
# Train and Evaluate the Model
|
| 95 |
+
def train_model(model, criterion, optimizer, num_epochs=5):
|
| 96 |
+
best_model_wts = copy.deepcopy(model.state_dict())
|
| 97 |
+
best_acc = 0.0
|
| 98 |
+
train_loss_history = []
|
| 99 |
+
val_loss_history = []
|
| 100 |
+
train_acc_history = []
|
| 101 |
+
val_acc_history = []
|
| 102 |
+
|
| 103 |
+
for epoch in range(num_epochs):
|
| 104 |
+
st.write(f'Epoch {epoch+1}/{num_epochs}')
|
| 105 |
+
st.write('-' * 10)
|
| 106 |
+
|
| 107 |
+
for phase in ['train', 'val']:
|
| 108 |
+
if phase == 'train':
|
| 109 |
+
model.train()
|
| 110 |
+
else:
|
| 111 |
+
model.eval()
|
| 112 |
+
|
| 113 |
+
running_loss = 0.0
|
| 114 |
+
running_corrects = 0
|
| 115 |
+
|
| 116 |
+
for inputs, labels in dataloaders[phase]:
|
| 117 |
+
inputs = inputs.to(device)
|
| 118 |
+
labels = labels.to(device)
|
| 119 |
+
|
| 120 |
+
optimizer.zero_grad()
|
| 121 |
+
|
| 122 |
+
with torch.set_grad_enabled(phase == 'train'):
|
| 123 |
+
outputs = model(inputs)
|
| 124 |
+
_, preds = torch.max(outputs, 1)
|
| 125 |
+
loss = criterion(outputs, labels)
|
| 126 |
+
|
| 127 |
+
if phase == 'train':
|
| 128 |
+
loss.backward()
|
| 129 |
+
optimizer.step()
|
| 130 |
+
|
| 131 |
+
running_loss += loss.item() * inputs.size(0)
|
| 132 |
+
running_corrects += torch.sum(preds == labels.data)
|
| 133 |
+
|
| 134 |
+
epoch_loss = running_loss / len(dataloaders[phase].dataset)
|
| 135 |
+
epoch_acc = running_corrects.double() / len(dataloaders[phase].dataset)
|
| 136 |
+
|
| 137 |
+
if phase == 'train':
|
| 138 |
+
train_loss_history.append(epoch_loss)
|
| 139 |
+
train_acc_history.append(epoch_acc)
|
| 140 |
+
else:
|
| 141 |
+
val_loss_history.append(epoch_loss)
|
| 142 |
+
val_acc_history.append(epoch_acc)
|
| 143 |
+
|
| 144 |
+
st.write(f'{phase} Loss: {epoch_loss:.4f} Acc: {epoch_acc:.4f}')
|
| 145 |
+
|
| 146 |
+
if phase == 'val' and epoch_acc > best_acc:
|
| 147 |
+
best_acc = epoch_acc
|
| 148 |
+
best_model_wts = copy.deepcopy(model.state_dict())
|
| 149 |
+
|
| 150 |
+
model.load_state_dict(best_model_wts)
|
| 151 |
+
|
| 152 |
+
# Plot training history
|
| 153 |
+
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 5))
|
| 154 |
+
ax1.plot(train_loss_history, label='Training Loss')
|
| 155 |
+
ax1.plot(val_loss_history, label='Validation Loss')
|
| 156 |
+
ax1.legend(loc='upper right')
|
| 157 |
+
ax1.set_title('Training and Validation Loss')
|
| 158 |
+
|
| 159 |
+
ax2.plot(train_acc_history, label='Training Accuracy')
|
| 160 |
+
ax2.plot(val_acc_history, label='Validation Accuracy')
|
| 161 |
+
ax2.legend(loc='lower right')
|
| 162 |
+
ax2.set_title('Training and Validation Accuracy')
|
| 163 |
+
|
| 164 |
+
st.pyplot(fig)
|
| 165 |
+
|
| 166 |
+
return model
|
| 167 |
+
|
| 168 |
+
if st.button('Train Model'):
|
| 169 |
+
model_ft = train_model(model_ft, criterion, optimizer_ft, num_epochs)
|
| 170 |
+
# Save the Model
|
| 171 |
+
torch.save(model_ft.state_dict(), 'fine_tuned_resnet.pth')
|
| 172 |
+
st.write("Model saved as 'fine_tuned_resnet.pth'")
|
| 173 |
+
|