DeepLearning_Model_Evaluators

This repository demonstrates how I evaluate and test a Computer Vision (CV) face & gender detection model without a GPU, using Google Colab webcam integration.

The project combines:

• Haar Cascade Face Detection
• Custom Gender Detection Model (MobileNetV3 Small – V13)

It is especially useful for users who do not have a GPU-based PC but still want to test deep learning models in real time.

---

🚀 Project Overview

• Face Detection:
  Uses OpenCV’s haarcascade_frontalface_default.xml for real-time face detection.

• Gender Classification:
  A custom-trained MobileNetV3 Small model (binary classification: Male / Female).

• Platform:
  Google Colab (Webcam-based testing via JavaScript bridge).

---

🧠 Model Details

Component	Description
Face Detector	Haar Cascade (OpenCV)
Gender Model	MobileNetV3 Small (V13)
Input Size	224 × 224
Output	Sigmoid (Male / Female)
Framework	TensorFlow / Keras

---

📊 Results

✅ Accuracy

• Face capture accuracy: 75% – 95%
• Works well for single-face, front-facing detection

⚠️ Known Issues

• Face tracking is not perfectly smooth
• Multiple face detection needs improvement
• Accuracy drops when the face is:
  • Too far from the camera
  • Partially visible
  • Poorly lit

---

🧪 Test Scenarios Covered

• [yes] Single face detection
• [yes] Gender prediction
• [yes] Multiple face detection (needs improvement)
• [yes] Long-distance face detection
• [yes] Face mismatch edge cases

---

🛠 Installation (Google Colab)

Run the following command in Colab:

!pip install tensorflow opencv-python-headless pillow

The goal is that make it more perfect Model because testing work well........................ Thank you!