[Feature Added] > frame_detector. This detects frames and spits out AI or REAL, based on the provided Video. Tested completely, with EDGE CASES.

.gitignore

@@ -36,4 +36,5 @@ folder_struct.txt
 # Misc
 docs/
 AGENTS.md
-to_be_implemented.md
+to_be_implemented.md
+Testing

backend/app/ai/video/frame_detector.py

@@ -0,0 +1,70 @@
+import torch
+import torch.nn as nn
+from torchvision import models, transforms
+from PIL import Image
+import numpy as np
+import logging
+
+logger = logging.getLogger(__name__)
+
+class FrameDetector:
+    def __init__(self):
+        self.device = torch.device("cpu")
+
+        try:
+            self.model = models.efficientnet_b0(weights=models.EfficientNet_B0_Weights.IMAGENET1K_V1)
+
+            for param in self.model.parameters():
+                param.requires_grad = False
+
+            num_ftrs = self.model.classifier[1].in_features
+            self.model.classifier[1] = nn.Linear(num_ftrs, 1)
+
+            self.model = self.model.to(self.device)
+            self.model.eval()
+
+            self.transforms = transforms.Compose([
+                transforms.Resize((224, 224)),
+                transforms.ToTensor(),
+                transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
+            ])
+            logger.info("EfficientNet-B0 CPU Inference Engine Initialized Successfully.")
+
+        except Exception as e:
+            logger.error(f"FATAL: Could not boot AI model: {str(e)}")
+            raise
+    
+    def predict(self, frame: np.ndarray) -> float:
+        try:
+            """
+                Flipping the frame matrix, since OpenCV reads
+                colors backwards (B, G, R). But PyTorch expects (R, G, B).
+                Hence the flipping of matrix
+            """
+
+            rgb_frame = frame[:, :, ::-1]
+            pil_image = Image.fromarray(rgb_frame)
+
+            input_tensor = self.transforms(pil_image).unsqueeze(0).to(self.device)
+
+            with torch.no_grad():
+                raw_score = self.model(input_tensor)
+                probability = torch.sigmoid(raw_score).item()
+            
+            return probability
+        
+        except Exception as e:
+            logger.error(f"AI Frame Prediction Crashed: {str(e)}")
+            return 0.0
+        
+try:
+    _detector_singleton = FrameDetector()
+except Exception:
+    _detector_singleton = None
+
+def analyze_frame(frame: np.ndarray) -> float:
+    if _detector_singleton is None:
+        logger.error("AI Engine is offline. Returning safe default.")
+        return 0.0
+    
+    return _detector_singleton.predict(frame)