🎭 Facial Emotion Recognition System

🎭 Facial Emotion Recognition System (Scikit-Learn Edition)

🎭 Facial Emotion Recognition System (FER-2013)

Author: Corey Leath (GitHub: Trojan3877)

Status: Production-Ready | L5/L6 Quality | Deployable | GPU-Accelerated

---

📘 Overview

This project is a full Facial Emotion Recognition (FER) system built using the FER-2013 dataset.
It includes:

• A complete training pipeline (preprocessing → model → metrics)
• A production-level CNN architecture
• Inference via predict.py
• A clean UI via Streamlit (streamlit_app.py)
• Full visualization (accuracy curves, confusion matrix)
• A professional ML engineering structure
• Deployment-ready components for portfolio use

GPU Deployment This model supports NVIDIA GPUs (CUDA 11.8, cuDNN 8.6) and was tested on [instance type] with latency of ~XX ms per image.

This project demonstrates Machine Learning Engineer (L5/L6) capabilities, including:

• Model design
• Training with callbacks
• Data validation
• Visualization
• Inference API
• Deployment readiness
• Reproducibility (requirements.txt)

---

🧠 Architecture

Facial Emotion Recognition System │ ├── src/ │ ├── train.py # Train CNN model (L6 quality) │ ├── predict.py # Run inference on images │ ├── streamlit_app.py # Web UI for predictions │ ├── visualize.py # Plots + confusion matrix │ ├── emotion_model_final.h5 # Saved trained model (if included) ├── fer2013.csv # Training dataset ├── history.json # Training history ├── requirements.txt # Reproducible environment └── README.md # You are here

flowchart TD

subgraph UI["User Interfaces"]
    ST[Streamlit App<br/>:8501]
    API_DOCS[FastAPI Docs<br/>/docs]
end

subgraph API["Inference & API Layer"]
    FASTAPI[FastAPI Server<br/>/predict]
    MODEL[Loaded FER Model<br/>emotion_model_final.h5]
end

subgraph PRE["Preprocessing"]
    CV2[OpenCV Image Processing<br/>48x48 Grayscale]
end

subgraph LOGS["MLOps & Logging"]
    MLFLOW[MLflow Tracking<br/>Metrics / Params / Artifacts]
    SNOWFLAKE[Snowflake Warehouse<br/>Prediction Logs]
end

subgraph STORAGE["Models & Artifacts"]
    MODEL_FILE[(emotion_model_final.h5)]
    HISTORY[(history.json)]
end

subgraph K8S["Kubernetes Cluster"]
    POD1((FER Pod #1))
    POD2((FER Pod #2))
    LB[LoadBalancer Service<br/>80 → 8000/8501]
end

subgraph DOCKER["Containerization"]
    DOCKERFILE[Docker Image<br/>FastAPI + Streamlit]
end

%% FLOW CONNECTIONS

ST --> FASTAPI
API_DOCS --> FASTAPI

FASTAPI --> PRE
PRE --> MODEL
MODEL --> FASTAPI

FASTAPI --> SNOWFLAKE
FASTAPI --> MLFLOW

MLFLOW --> HISTORY
MLFLOW --> MODEL_FILE

DOCKERFILE --> POD1
DOCKERFILE --> POD2

POD1 --> LB
POD2 --> LB
LB --> ST
LB --> FASTAPI

---

🚀 Features

✔ 7-class emotion detection

• Angry
• Disgust
• Fear
• Happy
• Sad
• Surprise
• Neutral

✔ L6-quality CNN model

• 3 convolutional blocks
• BatchNorm + Dropout for stability
• Final Dense classifier
• Adaptive learning rate
• EarlyStopping + Checkpointing

✔ Training visualizations

Saved in /plots/:

• accuracy_curve.png
• loss_curve.png
• confusion_matrix.png
• classification_report.txt

✔ Streamlit Web App

Run with: