Edit

Author: gjustin40

3. Model/classification/datasets.py

@@ -1,12 +1,13 @@
 import os
-from torchvision.datasets import CIFAR10, CIFAR100, MNIST
+from torchvision.datasets import CIFAR10, CIFAR100, MNIST, ImageFolder
 
 def get_dataset(name, train_transform=None, test_transform=None):
     data_paths = {
         'mnist': r'C:\Users\gjust\Documents\Github\data',
         'cifar10': r'C:\Users\gjust\Documents\Github\data',
         'cifar100': r'C:\Users\gjust\Documents\Github\data',
-        'coco': r'C:\Users\gjust\Documents\Github\data\COCO'
+        'coco': r'C:\Users\gjust\Documents\Github\data\COCO',
+        'catdog': r'C:\Users\gjust\Documents\Github\data\dogs-vs-cats'
     }
 
     data_path = data_paths[name]
@@ -30,8 +31,14 @@ def get_dataset(name, train_transform=None, test_transform=None):
         train_set = CIFAR100(root=data_path, transform=train_transform, train=True, download=True)
         test_set = CIFAR100(root=data_path, transform=test_transform, train=False, download=True)
 
+    elif name == 'catdog':
+        train_path = os.path.join(data_path, 'train')
+        # test_path = os.path.join(data_path, 'test')
+        train_set = ImageFolder(root=train_path, transform=train_transform)
+        test_set = ImageFolder(root=train_path, transform=test_transform)
+        
     return train_set, test_set
 
 if __name__ == '__main__':
-    data_path = get_dataset('cifar10')
+    data_path = get_dataset('catdog')
     print(data_path)

3. Model/classification/metrics.py

@@ -0,0 +1,36 @@
+import torchmetrics
+
+if __name__ == '__main__':
+    import torch
+    # import our library
+    import torchmetrics
+
+    # initialize metric
+    metric = torchmetrics.Accuracy(num_classes=5).cuda()
+    recall = torchmetrics.Recall(num_classes=5, multiclass=True, average='samples').cuda()
+    precision = torchmetrics.Precision(num_classes=5, multiclass=True).cuda()
+
+    n_batches = 100
+    for i in range(n_batches):
+        # simulate a classification problem
+        preds = torch.randn(10, 5).cuda()
+        target = torch.randint(5, (10,)).cuda()
+        
+        # metric on current batch
+        metric(preds, target)
+        recall(preds, target)
+        precision(preds, target)
+        result(preds, target)
+        #print(f"Accuracy on batch {i}: {acc:0.4f}")
+    # metric on all batches using custom accumulation
+    acc = metric.compute()
+    recall = metric.compute()
+    precision = metric.compute()
+
+    print(f"Accuracy on all data: {acc:0.4}")
+    print(f"Recall: {recall:0.4}")
+    print(f"Precision: {precision:0.4}")
+    # print(f'result : {result.compute()}')
+    acc = metric.compute()
+    # Reseting internal state such that metric ready for new data
+    metric.reset()

3. Model/classification/models.py

@@ -2,22 +2,36 @@
 from ResNet.model import ResNet, BasicBlock, Bottleneck
 from GoogLeNet.model import GoogLeNet
 
+import torchvision
 
-def get_model(name='resnet18', num_classes=10):
-    models = {
-        
-        'vgg11': VGG('vgg11', num_classes=num_classes),
-        'vgg13': VGG('vgg13', num_classes=num_classes),
-        'vgg16': VGG('vgg16', num_classes=num_classes),
-        'vgg19': VGG('vgg19', num_classes=num_classes),
-        
-        'resnet18': ResNet(BasicBlock, [2, 2, 2, 2], num_classes=num_classes),
-        'resnet34': ResNet(BasicBlock, [3, 4, 6, 3], num_classes=num_classes),
-        'resnet50': ResNet(Bottleneck, [3, 4, 6, 3], num_classes=num_classes),
-        'resnet101': ResNet(Bottleneck, [3, 4, 23, 3], num_classes=num_classes),
-        'resnet152': ResNet(Bottleneck, [3, 8, 36, 3], num_classes=num_classes),
+
+def get_model(name='resnet18', num_classes=10, pretrained=False):
+    
+    if pretrained:
+        models = {
+            'vgg11': torchvision.models.vgg11_bn(pretrained=True),
+            'vgg13': torchvision.models.vgg13_bn(pretrained=True),
+            'vgg16': torchvision.models.vgg16_bn(pretrained=True),
+            'vgg19': torchvision.models.vgg19_bn(pretrained=True)
+        }
 
-        'googlenet': GoogLeNet(num_classes=num_classes)
-    }
+        return models[name]
+    
+    else:
+        models = {         
+            
+            'vgg11': VGG('vgg11', num_classes=num_classes),
+            'vgg13': VGG('vgg13', num_classes=num_classes),
+            'vgg16': VGG('vgg16', num_classes=num_classes),
+            'vgg19': VGG('vgg19', num_classes=num_classes),
+            
+            'resnet18': ResNet(BasicBlock, [2, 2, 2, 2], num_classes=num_classes),
+            'resnet34': ResNet(BasicBlock, [3, 4, 6, 3], num_classes=num_classes),
+            'resnet50': ResNet(Bottleneck, [3, 4, 6, 3], num_classes=num_classes),
+            'resnet101': ResNet(Bottleneck, [3, 4, 23, 3], num_classes=num_classes),
+            'resnet152': ResNet(Bottleneck, [3, 8, 36, 3], num_classes=num_classes),
+            
+            'googlenet': GoogLeNet(num_classes=num_classes)
+        }
 
-    return models[name]
+        return models[name] 

3. Model/classification/remove.py

@@ -0,0 +1,3 @@
+import torch
+
+print(torch.randint(5, (10,)))