Siddhartha276/Fall_Detection

Fall Detection Model using EfficientNetB0

This model detects whether a person has fallen in an input image using transfer learning with EfficientNetB0. It is trained for binary classification: Fall Detected or No Fall Detected.

---

Model Architecture

• Base Model: EfficientNetB0 (include_top=False, pretrained on ImageNet)
• Top Layers:
  • GlobalAveragePooling2D
  • BatchNormalization
  • Dropout (0.4)
  • Dense (sigmoid activation)
• Loss Function: Binary Crossentropy
• Optimizer: Adam

The model was trained in two phases:

• Initial training with base model frozen (10 epochs)
• Fine-tuning with selective unfreezing (5 additional epochs)

Data augmentation techniques like RandomFlip, RandomRotation, and RandomZoom are used during training.

---

The repository contains two versions of the model:

1. Keras .h5 model
   • Full model for general use on machines with standard computational capacity.
2. TensorFlow Lite .tflite model
   • Optimized for mobile and edge devices with limited computing power.

---

How to Use

1. Load the Model from Hugging Face

from huggingface_hub import from_pretrained_keras

# Replace with your actual repo path
model = from_pretrained_keras("author-username/model-name")

2. Run Inference on an Image

from tensorflow.keras.preprocessing import image
from tensorflow.keras.applications.efficientnet import preprocess_input
import numpy as np
import matplotlib.pyplot as plt

# Define image size
IMG_SIZE = (224, 224)

# Load and preprocess the image
img_path = "image_uri" # Your image uri (from the drive or local storage)
img = image.load_img(img_path, target_size=IMG_SIZE)
img_array = image.img_to_array(img)
img_array = np.expand_dims(img_array, axis=0)
img_array = preprocess_input(img_array)

# Display the image
plt.imshow(img)
plt.axis("off")
plt.show()

# Make prediction
prediction = model.predict(img_array)
print(prediction)

# Interpret prediction
if prediction[0] < 0.15:
    print("Prediction: 🚨 Fall Detected! 🚨")
else:
    print("Prediction: ✅ No Fall Detected.")