Merge branch 'models' into develop

Feat: Integrate improved BetterCNN architecture and ClearML training logic.
Fixes: Resolved conflicts in run_training.py related to model initialization.

.gitignore

@@ -1,10 +1,8 @@
-<<<<<<< HEAD
 .vscode/
 .venv/
 .vscode/
 .models/
 __pycache__/
-=======
 
 # Python environment
 venv/
@@ -18,4 +16,4 @@ __pycache__/
 
 # Generated files from data_preparation.py
 class_distribution.png
->>>>>>> 04cb88662062ef6b880c627546d067fa0cedfa8b
+

best_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23a4c08eaad4b40290eca84e6a8fa3e1d69bdf4312d5db6db5de96d1d8753024
+size 130261986

models/modelTwo.py

@@ -0,0 +1,65 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+class BetterCNN(nn.Module):
+    def __init__(self, noOfClasses=39):
+        super(BetterCNN, self).__init__()
+
+        # 32 Channels 
+        # We use padding=1 to keep spatial size same before pooling
+        self.conv1 = nn.Conv2d(in_channels=3, out_channels=32, kernel_size=3, padding=1)
+        self.bn1 = nn.BatchNorm2d(32)
+
+        # 64 Channels
+        self.conv2 = nn.Conv2d(32, 64, kernel_size=3, padding=1)
+        self.bn2 = nn.BatchNorm2d(64)
+
+        # 128 Channels      
+        self.conv3 = nn.Conv2d(64, 128, kernel_size=3, padding=1)
+        self.bn3 = nn.BatchNorm2d(128)
+
+        # 256 Channels
+        self.conv4 = nn.Conv2d(128, 256, kernel_size=3, padding=1)
+        self.bn4 = nn.BatchNorm2d(256)
+
+        # Pooling layer 
+        self.pool = nn.MaxPool2d(2, 2)
+
+        # Adaptive Pooling  
+        self.adaptive_pool = nn.AdaptiveAvgPool2d((4, 4))
+
+        # Classification Head
+        self.fc1 = nn.Linear(256 * 4 * 4, 1024)
+        self.dropout = nn.Dropout(0.5) # Dropout after Linear layer
+        
+        self.fc2 = nn.Linear(1024, 512)
+        self.fc3 = nn.Linear(512, noOfClasses)
+
+    def forward(self, x):
+        # Block 1
+        x = self.conv1(x)
+        x = self.bn1(x)     # BatchNorm 
+        x = F.relu(x)
+        x = self.pool(x)    
+
+        # Block 2
+        x = self.pool(F.relu(self.bn2(self.conv2(x))))
+
+        # Block 3
+        x = self.pool(F.relu(self.bn3(self.conv3(x))))
+
+        # Block 4
+        x = self.pool(F.relu(self.bn4(self.conv4(x))))
+        
+        # Adapt & Flatten
+        x = self.adaptive_pool(x) 
+        x = torch.flatten(x, 1)   # Flattens to (Batch, 4096)
+        
+        # Dense Layers
+        x = F.relu(self.fc1(x))
+        x = self.dropout(x)       # Regularization
+        x = F.relu(self.fc2(x))
+        x = self.fc3(x)           # No activation needed here (handled by CrossEntropyLoss)
+
+        return x

subset_indices.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:972615a5b506b5ee2490f61866c26a4a2f9e2498c0baedb195a2a0d10a62e76f
+size 111016

trainingModel/run_training.py

@@ -3,7 +3,7 @@ from clearml import Task
 from dataPrep.helpers.clearml_data import extract_latest_data_task
 
 import torch
-from models.modelOne import modelOne
+from models.modelTwo import BetterCNN
 from trainingModel.helpers.Training import train_model
 
 
@@ -37,7 +37,7 @@ training_task.connect(training_config)
 
 
 # -------- Build the ML model --------
-model = modelOne(noOfClasses=training_config["num_classes"])
+model = BetterCNN(noOfClasses=training_config["num_classes"])
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 model.to(device)