SignVision

📊 Project Results & Performance

After training for 10 Epochs, the model achieved a Validation Accuracy of 97.93%.

Training Metrics

Epoch 10 Loss: 192.30 Accuracy: 97.93%

Confusion Matrix

The confusion matrix below visualizes the performance of the classification algorithm. As seen in the diagonal density, the model predicts the vast majority of classes correctly.

Classification Report

Detailed precision, recall, and F1-scores for all classes:

          precision    recall  f1-score   support

       A       1.00      1.00      1.00         1
       B       1.00      1.00      1.00         1
       C       1.00      1.00      1.00         1
       D       1.00      1.00      1.00         1
       ...     ...       ...       ...        ...
       X       1.00      1.00      1.00         1
       Y       1.00      1.00      1.00         1
       Z       1.00      1.00      1.00         1
     del       0.00      0.00      0.00         0
 nothing       1.00      1.00      1.00         1
   space       1.00      1.00      1.00         1

accuracy                           1.00        28

macro avg 0.97 0.97 0.97 28 weighted avg 1.00 1.00 1.00 28

---

🧠 Model Interpretability

To understand how the CNN makes decisions, we utilized visualization techniques to peek inside the "black box."

Activation Maps

The image below shows the output of the second convolutional layer (conv2). These maps highlight the specific features (edges, curves, textures) the network is detecting at this stage.

Grad-CAM (Class Activation Mapping)

We used Grad-CAM to visualize where the model "looks" when making a prediction. The heatmap overlays the original image, showing the regions of the hand that contributed most to the classification decision.

---

SYLLABUS & RESOURCES

Below is the curriculum followed to build this project, ranging from Python basics to Model Interpretation.

WEEK 0: Installation of Python and Anaconda

Python

• Getting Started with Python in VS Code (Official Video)
• Getting Started with Python in VS Code

Anaconda and Jupyter Notebook

• How to Install Anaconda and Jupyter Notebook on Windows 11
• Install Anaconda Python, Jupyter Notebook And Spyder on Mac
• Install Anaconda Python, Jupyter Notebook, Spyder on Ubuntu 22.04 LTS Linux

---

WEEK 1–2: Python, NumPy, Pandas, Matplotlib, Git, GitHub

Python

Video-Based - Python Tutorial

Text-Based - Python Tutorial | W3Schools

Hindi Resources - Python Tutorial for Beginners

Official Documentation

• Python Documentation

NumPy

Video-Based - Python NumPy Tutorial for Beginners

Text-Based - NumPy Tutorial (GitHub)

• NumPy Tutorial | W3Schools

Hindi Resources - Numpy Tutorial in Hindi

Official Documentation

• Numpy Documentation

Pandas

Video-Based - Complete Python Pandas Data Science Tutorial

Text-Based - Pandas Data Science Tutorial (GitHub)

• Pandas Tutorial | W3Schools

Hindi Resources - Python Pandas Tutorial in Hindi

Official Documentation

• Pandas Documentation

Matplotlib

Video-Based - Matplotlib Crash Course

Text-Based - Matplotlib Cheatsheets — Visualization with Python

• Matplotlib Tutorial | W3Schools

Hindi Resources - Python Matplotlib Tutorial in Hindi

Official Documentation

• Matplotlib Documentation

Git and GitHub

Video-Based - Git and GitHub for Beginners - Crash Course

Text-Based - Git Tutorial

Hindi Resources - Complete Git and GitHub Tutorial for Beginners

Official Documentation

• Git and Github Documentation

---

📌 Assignment 1 (End of Week 2)

---

WEEK 3: Linear and Logistic Regression

Linear Regression

Video-Based - Linear Regression Algorithm | Edureka

Text-Based - Python Machine Learning – Linear Regression

Hindi Resources - Linear Regression Implementation | Machine Learning in Hindi

• Linear Regression with Examples & Calculations

Logistic Regression

Video-Based - Logistic Regression in Python | Edureka

Text-Based - Python Machine Learning – Logistic Regression

Hindi Resources - Logistic Regression with Example

---

📌 Assignment 2 (End of Week 3)

---

WEEK 4-5: Neural Networks and Convolutional Neural Networks

Video-Based - Deep Learning Crash Course for Beginners

• Wix Studio

Text-Based - CNN Explainer

• PyTorch Tutorial

Hindi Resources - What is Convolutional Neural Network (CNN)

• PyTorch for Beginners

---

📌 Assignment 3 (End of Week 4)

---

📌 Assignment 4 (End of Week 5)

---

WEEK 6: Model Interpretation — Grad-CAM

Video-Based - Visualizing CAMs

• Grad-CAM with Python

Text-Based - Grad-CAM for Explaining Computer Vision Models

• Grad-CAM In PyTorch
• Grad-CAM: Visual Explanations from Deep Networks

---

Acknowledgements

Parts of this project were adapted from WINTER-PROJECT-ACTIVE-LEARNING by Bhavishya Gupta.