File size: 2,102 Bytes
82b9fde
704940b
5e9813e
82b9fde
 
5e9813e
 
82b9fde
704940b
 
82b9fde
 
5e9813e
82b9fde
394fedc
 
82b9fde
394fedc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5e9813e
394fedc
 
 
 
 
 
 
 
 
 
 
 
 
5e9813e
394fedc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5e9813e
394fedc
5e9813e
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
---
title: SignLanguageClassification
emoji: 🀟
colorFrom: red
colorTo: red
sdk: streamlit
app_file: src/streamlit_app.py
pinned: false
short_description: Upload a hand sign image and the CNN predicts
license: mit
---

# 🀟 Sign Language Classification (CNN)

This project demonstrates a Convolutional Neural Network (CNN) for classifying sign language hand gestures from images.  
A lightweight CNN was trained on grayscale images and deployed as a Streamlit web application.

---

## πŸš€ Demo
Upload an image of a hand sign (JPG/PNG), and the model predicts the corresponding class along with a confidence score and top predictions.

---

## πŸ“Š Dataset
- Grayscale images with a resolution of **28Γ—28 pixels**
- Images are provided in **CSV format** (pixel values + labels)
- Some dataset versions contain **non-continuous label indices**
- An additional **"Unknown"** class is used to ensure consistency between labels and model outputs

---

## 🧠 Model
- Framework: **TensorFlow / Keras**
- Architecture: **Sequential CNN**
  - Convolution + Batch Normalization + MaxPooling layers
  - Fully connected layers with Dropout
- Loss: `sparse_categorical_crossentropy`
- Optimizer: Adam
- Training without GPU acceleration

---

## πŸ“ˆ Results
- Test Accuracy: **~94%**
- Strong performance across most classes
- Confusion matrix shows only minor misclassifications between visually similar hand gestures

---

## ⚠️ Limitations
Predictions may vary for real-world images due to background noise, lighting conditions, and differences from the training data, which mainly contains centered and cropped hand images.

---

## πŸ–₯️ Streamlit App
The Streamlit app allows users to:
- Upload an image (JPG/PNG)
- View the input image
- See the predicted class and confidence score
- Inspect the top prediction probabilities

---

## πŸ“ Project Files
The following files are required in the repository root:

app.py
signlanguage_cnn.keras
label_map.json
requirements.txt
README.md

yaml
Code kopieren

---

## ▢️ Run Locally
```bash
pip install -r requirements.txt
streamlit run app.py