CNNs-for-Computer-Vision

This is my attempt to use a convolutional neural network multi-output model that classifies images of people based on their ethnicity and predicts their age and gender.

The project explores multi-task learning, where a shared CNN backbone learns general facial features and three output heads specialise in classification (ethnicity and gender) and regression (age prediction).

Key features

• Robust Data Pipeline - Engineered a tf.data pipeline with safe augmentation techniques.
• Single backbone + three prediction heads:
  • Age (Regression)
  • Ethnicity (Classification)
  • Gender (Classification)
• Transfer Learning Comparison: Benchmarked the custom model against a fine-tuned MobileNetV2.
• Streamlit Application — Real-time face attribute prediction via image upload or live webcam.

Dataset: https://www.kaggle.com/datasets/moritzm00/utkface-cropped

Model Overview

The final model uses a custom VGG-style architecture with Global Average Pooling to reduce parameter bloat and improve generalization.

Task	Output	Metric	Best Validation
Age	Regression	MAE	0.0749
Gender	Classification	Accuracy	90.11%
Ethnicity	Classification	Accuracy	75.45%

Download model.keras from Google Drive Later, the custom model was used to build a Streamlit application that predicts the three demographics.

View Full Training Notebook

For full training details — including model code, metrics, plots, and experiments, see: CNN_tf_keras.ipynb (located in the repo root)

Streamlit App Features

• Upload an image and get face attribute predictions
• Live webcam inference using Streamlit_WebRTC
• Lightweight & CPU-friendly deployment (uses lazy loading)
• Handles transparent (RGBA) images

Demo Output

Below is an example of the face-analysis prediction generated by the Streamlit app:

Installation

Follow the steps below to launch the Face Analysis Streamlit application on your local machine.

Prerequisites Make sure you have:

• Python 3.10
• Git
• OS: Works on Windows / macOS / Linux

1. Clone the Repository

git clone https://github.com/jatin-reddy/CNNs-for-Computer-Vision.git
cd CNNs-for-Computer-Vision

2. Create a Virtual Environment

macOS / Linux

python3.10 -m venv venv
source venv/bin/activate

Windows

python -m venv venv
venv\Scripts\activate

3. Install Dependencies

pip install --upgrade pip
pip install -r requirements.txt

4. Connect the app to the model

Download model.keras from Google Drive And then use the correct path in load_path() function on each page

5. Launch the Application

streamlit run Home.py

⚠ Disclaimer

Predictions are probabilistic and may not always be accurate.
Results should not be used for sensitive or critical decision-making.