util files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+ResUNet-segModel-weights.hdf5 filter=lfs diff=lfs merge=lfs -text

ResUNet-segModel-weights.hdf5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21efee8add11e7881172be0602d7e039271eb98bdaf7b085aa9546a2c710e424
+size 14961712

ResUNet.py

@@ -0,0 +1,100 @@
+import os
+import numpy as np
+import pandas as pd
+import cv2
+
+import tensorflow as tf
+from tensorflow.python.keras import Sequential
+from tensorflow.keras import layers, optimizers
+from tensorflow.keras.layers import *
+from tensorflow.keras.models import Model
+from tensorflow.keras.initializers import glorot_uniform
+from tensorflow.keras.utils import plot_model
+from tensorflow.keras.callbacks import ReduceLROnPlateau, EarlyStopping, ModelCheckpoint, LearningRateScheduler
+import tensorflow.keras.backend as K
+
+
+# lets create model now
+def resblock(X, f):
+    '''
+    function for creating res block
+    '''
+    X_copy = X  #copy of input
+
+    # main path
+    X = Conv2D(f, kernel_size=(1,1), kernel_initializer='he_normal')(X)
+    X = BatchNormalization()(X)
+    X = Activation('relu')(X)
+
+    X = Conv2D(f, kernel_size=(3,3), padding='same', kernel_initializer='he_normal')(X)
+    X = BatchNormalization()(X)
+
+    # shortcut path
+    X_copy = Conv2D(f, kernel_size=(1,1), kernel_initializer='he_normal')(X_copy)
+    X_copy = BatchNormalization()(X_copy)
+
+    # Adding the output from main path and short path together
+    X = Add()([X, X_copy])
+    X = Activation('relu')(X)
+
+    return X
+
+def upsample_concat(x, skip):
+    '''
+    funtion for upsampling image
+    '''
+    X = UpSampling2D((2,2))(x)
+    merge = Concatenate()([X, skip])
+
+    return merge
+
+
+def load_model():
+
+    input_shape = (256,256,3)
+    X_input = Input(input_shape) #iniating tensor of input shape
+
+    # Stage 1
+    conv_1 = Conv2D(16, 3, activation='relu', padding='same', kernel_initializer='he_normal')(X_input)
+    conv_1 = BatchNormalization()(conv_1)
+    conv_1 = Conv2D(16, 3, activation='relu', padding='same', kernel_initializer='he_normal')(conv_1)
+    conv_1 = BatchNormalization()(conv_1)
+    pool_1 = MaxPool2D((2,2))(conv_1)
+
+    # stage 2
+    conv_2 = resblock(pool_1, 32)
+    pool_2 = MaxPool2D((2,2))(conv_2)
+
+    # Stage 3
+    conv_3 = resblock(pool_2, 64)
+    pool_3 = MaxPool2D((2,2))(conv_3)
+
+    # Stage 4
+    conv_4 = resblock(pool_3, 128)
+    pool_4 = MaxPool2D((2,2))(conv_4)
+
+    # Stage 5 (bottle neck)
+    conv_5 = resblock(pool_4, 256)
+
+    # Upsample Stage 1
+    up_1 = upsample_concat(conv_5, conv_4)
+    up_1 = resblock(up_1, 128)
+
+    # Upsample Stage 2
+    up_2 = upsample_concat(up_1, conv_3)
+    up_2 = resblock(up_2, 64)
+
+    # Upsample Stage 3
+    up_3 = upsample_concat(up_2, conv_2)
+    up_3 = resblock(up_3, 32)
+
+    # Upsample Stage 4
+    up_4 = upsample_concat(up_3, conv_1)
+    up_4 = resblock(up_4, 16)
+
+    # final output
+    out = Conv2D(1, (1,1), kernel_initializer='he_normal', padding='same', activation='sigmoid')(up_4)
+
+    seg_model = Model(X_input, out)
+
+    return seg_model

eff.py

@@ -0,0 +1,145 @@
+import os
+import cv2
+import torch
+import torchvision
+from torch import nn
+from torchvision import transforms
+from PIL import Image
+
+class_names = ["Glioma Tumor", "Meningitis Tumor", "No Tumor", "Pituitary Tumor"]
+
+class CFG:
+    DEVICE = 'cpu'
+    NUM_DEVICES = torch.cuda.device_count()
+    NUM_WORKERS = os.cpu_count()
+    NUM_CLASSES = 4
+    EPOCHS = 16
+    BATCH_SIZE = 32
+    LR = 0.001
+    APPLY_SHUFFLE = True
+    SEED = 768
+    HEIGHT = 224
+    WIDTH = 224
+    CHANNELS = 3
+    IMAGE_SIZE = (224, 224, 3)
+
+
+class EfficientNetV2Model(nn.Module):
+    def __init__(self, backbone_model, name='efficientnet-v2-large',
+                 num_classes=CFG.NUM_CLASSES, device=CFG.DEVICE):
+        super(EfficientNetV2Model, self).__init__()
+
+        self.backbone_model = backbone_model
+        self.device = device
+        self.num_classes = num_classes
+        self.name = name
+
+        classifier = nn.Sequential(
+            nn.Flatten(),
+            nn.Dropout(p=0.2, inplace=True),
+            nn.Linear(in_features=1280, out_features=256, bias=True),
+            nn.GELU(),
+            nn.Dropout(p=0.2, inplace=True),
+            nn.Linear(in_features=256, out_features=num_classes, bias=False)
+        ).to(device)
+
+        self._set_classifier(classifier)
+
+    def _set_classifier(self, classifier:nn.Module) -> None:
+        self.backbone_model.classifier = classifier
+
+    def forward(self, image):
+        return self.backbone_model(image)
+
+
+def get_effiecientnetv2_model(
+    device: torch.device=CFG.NUM_CLASSES) -> nn.Module:
+    # Set the manual seeds
+    torch.manual_seed(CFG.SEED)
+    torch.cuda.manual_seed(CFG.SEED)
+
+    # Get model weights
+    model_weights = (
+        torchvision
+        .models
+        .EfficientNet_V2_L_Weights
+        .DEFAULT
+    )
+
+    # Get model and push to device
+    model = (
+        torchvision.models.efficientnet_v2_l(
+            weights=model_weights
+        )
+    ).to(device)
+
+    # Freeze Model Parameters
+    for param in model.features.parameters():
+        param.requires_grad = False
+
+    return model
+
+
+# Get EfficientNet v2 model
+backbone_model = get_effiecientnetv2_model(CFG.DEVICE)
+
+efficientnetv2_params = {
+    'backbone_model'    : backbone_model,
+    'name'              : 'efficientnet-v2-large',
+    'device'            : CFG.DEVICE
+}
+
+# Generate Model
+efficientnet_model = EfficientNetV2Model(**efficientnetv2_params)
+
+efficientnet_model.load_state_dict(
+    torch.load('efficientnetV2.pth', map_location=torch.device('cpu'))
+)
+
+
+# def predict_eff(image_path):
+#     # Define the image transformation
+#     transform = transforms.Compose([
+#         transforms.Resize((224, 224)),
+#         transforms.ToTensor()
+#     ])
+
+#     # Load and preprocess the image
+#     # image_path = "glioma.jpg"  # Replace with the path to your image
+#     image = Image.open(image_path)
+#     input_tensor = transform(image)
+#     input_batch = input_tensor.unsqueeze(0).to("cuda")  # Add batch dimension
+
+#     # Perform inference
+#     with torch.no_grad():
+#         output = efficientnet_model(input_batch).to("cuda")
+
+#     # You can now use the 'output' tensor as needed (e.g., get predictions)
+#     # print(output)
+#     return output
+
+
+def predict_eff(image_path):
+    # Define the image transformation
+    transform = transforms.Compose([
+        transforms.Resize((224, 224)),
+        transforms.ToTensor()
+    ])
+
+    # Load and preprocess the image
+    # image_path = "glioma.jpg"  # Replace with the path to your image
+    image = Image.open(image_path)
+    input_tensor = transform(image)
+    input_batch = input_tensor.unsqueeze(0).to(CFG.DEVICE)  # Add batch dimension
+
+    # Perform inference
+    with torch.no_grad():
+        output = efficientnet_model(input_batch).to(CFG.DEVICE)
+
+    # You can now use the 'output' tensor as needed (e.g., get predictions)
+    # print(output)
+    res = torch.softmax(output, dim=1)
+
+    probs = {class_names[i]: float(res[0][i]) for i in range(len(class_names))}
+
+    return probs

vit.py

@@ -0,0 +1,109 @@
+import os
+import torch
+import torchvision
+from torch import nn
+from torchvision import transforms
+from PIL import Image
+
+class CFG:
+    DEVICE = 'cpu'
+    NUM_DEVICES = torch.cuda.device_count()
+    NUM_WORKERS = os.cpu_count()
+    NUM_CLASSES = 4
+    EPOCHS = 16
+    BATCH_SIZE = 32
+    LR = 0.001
+    APPLY_SHUFFLE = True
+    SEED = 768
+    HEIGHT = 224
+    WIDTH = 224
+    CHANNELS = 3
+    IMAGE_SIZE = (224, 224, 3)
+    
+
+class VisionTransformerModel(nn.Module):
+    def __init__(self, backbone_model, name='vision-transformer',
+                 num_classes=CFG.NUM_CLASSES, device=CFG.DEVICE):
+        super(VisionTransformerModel, self).__init__()
+
+        self.backbone_model = backbone_model
+        self.device = device
+        self.num_classes = num_classes
+        self.name = name
+
+        self.classifier = nn.Sequential(
+            nn.Flatten(),
+            nn.Dropout(p=0.2, inplace=True),
+            nn.Linear(in_features=1000, out_features=256, bias=True),
+            nn.GELU(),
+            nn.Dropout(p=0.2, inplace=True),
+            nn.Linear(in_features=256, out_features=num_classes, bias=False)
+        ).to(device)
+
+    def forward(self, image):
+        vit_output = self.backbone_model(image)
+        return self.classifier(vit_output)
+    
+
+def get_vit_b32_model(
+    device: torch.device=CFG.NUM_CLASSES) -> nn.Module:
+    # Set the manual seeds
+    torch.manual_seed(CFG.SEED)
+    torch.cuda.manual_seed(CFG.SEED)
+
+    # Get model weights
+    model_weights = (
+        torchvision
+        .models
+        .ViT_L_32_Weights
+        .DEFAULT
+    )
+
+    # Get model and push to device
+    model = (
+        torchvision.models.vit_l_32(
+            weights=model_weights
+        )
+    ).to(device)
+
+    # Freeze Model Parameters
+    for param in model.parameters():
+        param.requires_grad = False
+
+    return model
+
+# Get ViT model
+vit_backbone = get_vit_b32_model(CFG.DEVICE)
+
+vit_params = {
+    'backbone_model'    : vit_backbone,
+    'name'              : 'ViT-L-B32',
+    'device'            : CFG.DEVICE
+}
+
+# Generate Model
+vit_model = VisionTransformerModel(**vit_params)
+
+vit_model.load_state_dict(
+    torch.load('vit_model.pth', map_location=torch.device('cpu'))
+)
+
+
+# Define the image transformation
+transform = transforms.Compose([
+    transforms.Resize((224, 224)),
+    transforms.ToTensor()
+])
+
+
+def predict(image_path):
+    image = Image.open(image_path)
+    input_tensor = transform(image)
+    input_batch = input_tensor.unsqueeze(0).to(CFG.DEVICE)  # Add batch dimension
+
+    # Perform inference
+    with torch.no_grad():
+        output = vit_model(input_batch).to(CFG.DEVICE)
+
+    # You can now use the 'output' tensor as needed (e.g., get predictions)
+    return output