_____           _       __        ______                  _ __  _           
  / ___/__________(_)___  / /_      / ____/___  ____ _____  (_) /_(_)___  ____ 
  \__ \/ ___/ ___/ / __ \/ __/_____/ /   / __ \/ __ `/ __ \/ / __/ / __ \/ __ \
 ___/ / /__/ /  / / /_/ / /_/_____/ /___/ /_/ / /_/ / / / / / /_/ / /_/ / / / /
/____/\___/_/  /_/ .___/\__/      \____/\____/\__, /_/ /_/_/\__/_/\____/_/ /_/ 
                /_/                          /____/                            

Your handwriting knows how you feel. Now the model does too.

Depression. Anxiety. Stress. — three invisible forces that shape lives.
This project reads them from the strokes of a pen.

---

👁️ What Is This?

EmotionDetectionUsingHandwriting is a full machine learning pipeline that detects emotional states from online handwriting signals captured on digital tablets. Pen trajectory data — coordinates, pressure, velocity, tilt — gets rendered into grayscale images and fed into a CNN that classifies the writer's emotional state.

It uses the EMOTHAW dataset (129 subjects, 7 handwriting tasks each) and labels derived from the DASS-21 (Depression Anxiety Stress Scales) questionnaire.

  .svc file  ──▶  parse pen data  ──▶  render trajectory  ──▶  CNN  ──▶  emotion

🌐 Live Demo

👉 Web Application:
https://script-cognition.streamlit.app/

⚠️ Deployment Notice
This Streamlit demo is intended for interface visualization and workflow demonstration only.
Due to deployment constraints (model size and training requirements), the hosted version does not include fully trained models.

For accurate predictions and complete functionality, please run the application locally after training the models as described below.

---

---

✦ Features at a Glance

	Feature	Detail
🧠	Dual CNN Architectures	EmotionCNN (2.7M params) · ResNet18 transfer learning (11.2M params)
🎯	Three Emotion Dimensions	Depression · Anxiety · Stress — train each independently
📊	DASS-21 Labels	Binary (Low / High) or full 5-class severity
🔬	29+ Handwriting Features	Velocity, pressure, curvature, stroke count, pen-lift frequency
🌐	Streamlit Web App	Upload .svc files · switch emotions · live trajectory viz
🖥️	Powerful CLI	Train · predict · visualize · extract — all from the terminal
⚡	GPU Ready	CUDA support out of the box

---

📊 Dataset at a Glance

  ╔══════════════════════════════════════════════════════╗
  ║  EMOTHAW Dataset                                     ║
  ║                                                      ║
  ║  Subjects    ██████████████████████████████  129     ║
  ║  Tasks/user  ████████████████████           7        ║
  ║  Samples     ████████████████████████████   900+     ║
  ║  HW Features ████████████████████████████   29+      ║
  ║  Emotions    ██████████████                 3        ║
  ║                                                      ║
  ║  Format: .svc  →  X, Y, timestamp, pressure,         ║
  ║                    pen status, azimuth, altitude     ║
  ╚══════════════════════════════════════════════════════╝

Emotion targets (binary Low / High per DASS-21 thresholds):

• 🟣 Depression — Normal → Extremely Severe
• 🔵 Anxiety — Normal → Extremely Severe
• 🩷 Stress — Normal → Extremely Severe

---

🗂️ Project Structure

EmotionDetectionUsingHandwriting/
│
├── ⚙️  config.py                    central configuration
├── 🚀  main.py                      unified CLI entry point
├── 🌐  app.py                       Streamlit web interface
├── 🏃  run_app.py                   web app launcher helper
├── 📋  requirements.txt
│
├── 📂  preprocessing/
│   └──  svc_parser.py              .svc → NumPy arrays
│
├── 📂  data/
│   ├──  label_loader.py            DASS scores → class labels
│   └──  dataset.py                 PyTorch Dataset + DataLoader
│
├── 📂  features/
│   ├──  image_generator.py         trajectories → images
│   └──  signal_features.py         29+ handwriting features
│
├── 📂  models/
│   └──  cnn_model.py               EmotionCNN + EmotionResNet
│
├── 📂  training/
│   └──  train_cnn.py               training loop + early stopping
│
├── 📂  evaluation/
│   └──  evaluate.py                metrics, confusion matrix, plots
│
├── 📂  inference/
│   └──  predict.py                 run inference on new samples
│
├── 📂  utils/
│   └──  visualization.py           trajectory & dataset plots
│
└── 📂  outputs/
    ├──  models/                    saved checkpoints (.pth)
    ├──  plots/                     generated figures
    └──  image_cache/               cached trajectory images

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• EMOTHAW dataset at ../archive/DataEmothaw/
• CUDA (optional, for GPU acceleration)

Installation

# Clone the repo
git clone https://github.com/kshitiz-arc/EmotionDetectionUsingHandwriting.git
cd EmotionDetectionUsingHandwriting

# Install dependencies
pip install -r requirements.txt

Expected dataset layout:

../archive/DataEmothaw/
├── DASS_scores.xls
├── user00001/
├── user00002/
└── ...

---

🎮 Usage

1 · Train

# Single emotion
python main.py train --model cnn --emotion anxiety
python main.py train --model cnn --emotion depression
python main.py train --model cnn --emotion stress

# All three in one shot
python main.py train_all --model cnn

# Custom hyperparams
python main.py train --model resnet --emotion anxiety \
    --epochs 50 --batch_size 32 --lr 0.0001

Checkpoints save to outputs/models/ as best_cnn_anxiety.pth, etc.

---

2 · Launch the Web App

python run_app.py
# or
streamlit run app.py

The app lets you switch emotion targets, upload .svc files, see confidence scores, and explore dataset statistics — all without touching the terminal again.

---

3 · Predict from the CLI

python main.py predict --svc_path "path/to/handwriting.svc" --emotion anxiety

══════════════════════════════════════════════════════
  HANDWRITING EMOTION PREDICTION
══════════════════════════════════════════════════════

  Input file  : u00001s00001_hw00001.svc
  Data points : 1474
  Dimension   : Anxiety

──────────────────────────────────────────────────────
  PREDICTED   : LOW
  CONFIDENCE  : 84.12%
──────────────────────────────────────────────────────

  Low   ████████████████████████████  0.8412
  High  ████                          0.1588

---

4 · Visualize

python main.py visualize --statistics     # class distribution
python main.py visualize --trajectory 1   # specific user's trajectory
python main.py visualize --all            # everything

5 · Extract Features

python main.py extract --svc_path file.svc   # single file
python main.py extract --dataset              # full dataset

6 · Python API

from handwriting_emotion_detection import (
    load_all_svc_files, load_labels, get_dataloaders,
    EmotionCNN, extract_signal_features, plot_trajectory
)

samples  = load_all_svc_files(DATASET_ROOT)
labels   = load_labels(DASS_SCORES_PATH, target_emotion="anxiety")
train_loader, val_loader, test_loader = get_dataloaders(samples, labels)

model    = EmotionCNN(num_classes=2)
features = extract_signal_features(samples[0]["data"])

---

🏗️ Model Architectures

EmotionCNN — fast & lean

Input (224×224)
    │
    ▼
Conv Block ×4  (BatchNorm + ReLU + MaxPool + Dropout)
    │
    ▼
Fully Connected  →  Softmax  →  [Low, High]

Parameters : ~2.7M
Best for   : small datasets, fast iteration

EmotionResNet — powerful & accurate

Input (224×224)
    │
    ▼
ResNet18 backbone  (pretrained on ImageNet)
    │
    ▼
Fine-tuned FC head  →  Softmax  →  [Low, High]

Parameters : ~11.2M
Best for   : maximum accuracy, transfer learning

Training setup — Adam optimizer · ReduceLROnPlateau scheduler · Cross-Entropy with class weights · Early stopping · 70 / 15 / 15 train/val/test split

---

⚙️ Configuration

All tunable params live in config.py:

Parameter	Default	Description
TARGET_EMOTION	"anxiety"	depression · anxiety · stress
USE_BINARY	True	True = Low/High · False = 5-class
IMAGE_SIZE	224	Trajectory render resolution (px)
BATCH_SIZE	16	Training batch size
NUM_EPOCHS	30	Max training epochs
LEARNING_RATE	1e-4	Initial Adam LR
EARLY_STOPPING_PATIENCE	7	Patience before stopping

---

🔧 Troubleshooting

Problem	Fix
No trained model found	Run python main.py train --model cnn --emotion anxiety first
Dataset not found	Check config.py paths, ensure DASS_scores.xls exists
streamlit: not found	pip install streamlit
Out of GPU memory	Lower BATCH_SIZE in config.py or switch to CPU
Poor accuracy	More epochs · lower LR · switch to --model resnet

---

🤝 Contributing

git checkout -b feature/your-idea
# make changes, write tests, update docs
git commit -m "feat: your idea"
git push origin feature/your-idea
# open a Pull Request

Please follow PEP 8 and add docstrings to any new functions.

---

📚 References

• EMOTHAW Dataset — UCI ML Repository
• DASS-21 Questionnaire
• PyTorch Docs
• Streamlit Docs

---

📄 License

APACHE 2.0 — see LICENSE for details.

---

Built by kshitiz-arc

The pen is mightier than the word — and now, smarter too.