EnYa32 commited on
Commit
394fedc
·
verified ·
1 Parent(s): 817ed41

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +61 -14
README.md CHANGED
@@ -12,24 +12,71 @@ license: mit
12
 
13
  # 🤟 Sign Language Classification (CNN)
14
 
15
- This project uses a Convolutional Neural Network (CNN) built with Keras (Sequential API) to classify sign language hand gesture images.
 
16
 
17
- ## Demo
18
- Upload a hand sign image (JPG/PNG) and the app predicts the class with confidence scores.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
 
20
- ## Model & Data
21
- - Input: grayscale image resized to **28×28**
22
- - Output: class probabilities (softmax)
23
- - The dataset was provided as CSV (pixel values + labels).
 
 
 
 
 
 
 
 
 
24
 
25
- ## Files
26
- Place these files in the repository root:
27
- - `app.py`
28
- - `signlanguage_cnn.keras`
29
- - `label_map.json`
30
- - `requirements.txt`
 
 
 
 
 
 
 
 
 
31
 
32
- ## Run locally
33
  ```bash
34
  pip install -r requirements.txt
35
  streamlit run app.py
 
12
 
13
  # 🤟 Sign Language Classification (CNN)
14
 
15
+ This project demonstrates a Convolutional Neural Network (CNN) for classifying sign language hand gestures from images.
16
+ A lightweight CNN was trained on grayscale images and deployed as a Streamlit web application.
17
 
18
+ ---
19
+
20
+ ## 🚀 Demo
21
+ Upload an image of a hand sign (JPG/PNG), and the model predicts the corresponding class along with a confidence score and top predictions.
22
+
23
+ ---
24
+
25
+ ## 📊 Dataset
26
+ - Grayscale images with a resolution of **28×28 pixels**
27
+ - Images are provided in **CSV format** (pixel values + labels)
28
+ - Some dataset versions contain **non-continuous label indices**
29
+ - An additional **"Unknown"** class is used to ensure consistency between labels and model outputs
30
+
31
+ ---
32
+
33
+ ## 🧠 Model
34
+ - Framework: **TensorFlow / Keras**
35
+ - Architecture: **Sequential CNN**
36
+ - Convolution + Batch Normalization + MaxPooling layers
37
+ - Fully connected layers with Dropout
38
+ - Loss: `sparse_categorical_crossentropy`
39
+ - Optimizer: Adam
40
+ - Training without GPU acceleration
41
+
42
+ ---
43
+
44
+ ## 📈 Results
45
+ - Test Accuracy: **~94%**
46
+ - Strong performance across most classes
47
+ - Confusion matrix shows only minor misclassifications between visually similar hand gestures
48
 
49
+ ---
50
+
51
+ ## ⚠️ Limitations
52
+ 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.
53
+
54
+ ---
55
+
56
+ ## 🖥️ Streamlit App
57
+ The Streamlit app allows users to:
58
+ - Upload an image (JPG/PNG)
59
+ - View the input image
60
+ - See the predicted class and confidence score
61
+ - Inspect the top prediction probabilities
62
 
63
+ ---
64
+
65
+ ## 📁 Project Files
66
+ The following files are required in the repository root:
67
+
68
+ app.py
69
+ signlanguage_cnn.keras
70
+ label_map.json
71
+ requirements.txt
72
+ README.md
73
+
74
+ yaml
75
+ Code kopieren
76
+
77
+ ---
78
 
79
+ ## ▶️ Run Locally
80
  ```bash
81
  pip install -r requirements.txt
82
  streamlit run app.py