Upload 3 files

FeatureExtraction.py

@@ -0,0 +1,41 @@
+import cv2
+import numpy as np
+from tensorflow.keras.models import load_model
+from tensorflow.keras.applications.resnet import ResNet152
+from tensorflow.keras.layers import AveragePooling2D, Flatten
+from tensorflow.keras.models import Model
+from tqdm import tqdm
+import os
+
+class FeatureExtractor:
+    def __init__(self, img_shape, channels):
+        self.seq_length = 40  # Number of frames to process
+        self.height = img_shape[0]
+        self.width = img_shape[1]
+        self.channels = channels
+
+        # Load ResNet152 model without the top fully connected layer
+        self.base_model = ResNet152(include_top=False, input_shape=(224, 224, 3), weights='imagenet')
+
+        # Freeze the base model layers
+        for layer in self.base_model.layers:
+            layer.trainable = False
+
+        # Adding an Average Pooling layer followed by Flatten
+        self.op = self.base_model.output
+        self.x_model = AveragePooling2D((7, 7), name='avg_pool')(self.op)
+        self.x_model = Flatten()(self.x_model)
+
+        # Create the feature extraction model
+        self.model = Model(self.base_model.input, self.x_model)
+
+    def extract_feature(self, frames_buffer):
+        x_op = np.zeros((2048, 40))  # Shape (features_dim, seq_length)
+        for i in range(len(frames_buffer)):
+            x_t = frames_buffer[i]
+            x_t = cv2.resize(x_t, (224, 224))  # Resize each frame to the required input size
+            x_t = np.expand_dims(x_t, axis=0)  # Add batch dimension
+            x = self.model.predict(x_t)
+            x_op[:, i] = x
+
+        return x_op

Prediction.py

@@ -0,0 +1,24 @@
+import cv2
+import numpy as np
+from tensorflow.keras.models import load_model
+import os
+from FeatureExtraction import FeatureExtractor
+
+model = load_model('orignal_model_b32.h5')
+model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
+
+
+# Initialize the feature extractor
+feature_extractor = FeatureExtractor(img_shape=(224, 224), channels=3)
+
+def predict_fight(frames_buffer):
+    # Extract feature
+    features_sequence = feature_extractor.extract_feature(frames_buffer)
+
+    # Transpose the feature sequence to match the shape
+    features_sequence = np.transpose(features_sequence, (1, 0))  # From (2048, 40) to (40, 2048)
+    features_sequence = np.expand_dims(features_sequence, axis=0)  # Add batch dimension (1, 40, 2048)
+
+    # Predict
+    prediction = model.predict(features_sequence)
+    return prediction > 0.8  # Returning a boolean for fight detection

orignal_model_b32.h5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad403c2e5014e3fd7589d74b801e4494191b329bcbaaa11fbd8d38effe3cc133
+size 191265304