Spaces:
Sleeping
Sleeping
|
Download README.md from EnYa32/SignLanguageClassification: direct link, hf CLI and curl.
- Browser
- Download file 2.1 kB
-
https://huggingface.co/spaces/EnYa32/SignLanguageClassification/resolve/main/README.md
- Command line
-
hf download hf://spaces/EnYa32/SignLanguageClassification/README.md
-
curl -L -o README.md https://huggingface.co/spaces/EnYa32/SignLanguageClassification/resolve/main/README.md
2.1 kB
| title: SignLanguageClassification | |
| emoji: π€ | |
| 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 | |
| # π€ 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. | |
| --- | |
| ## π 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. | |
| --- | |
| ## π Dataset | |
| - Grayscale images with a resolution of **28Γ28 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 | |
| --- | |
| ## π§ 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 | |
| --- | |
| ## π Results | |
| - Test Accuracy: **~94%** | |
| - Strong performance across most classes | |
| - Confusion matrix shows only minor misclassifications between visually similar hand gestures | |
| --- | |
| ## β οΈ 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. | |
| --- | |
| ## π₯οΈ 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 | |
| --- | |
| ## π 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 | |
| --- | |
| ## βΆοΈ Run Locally | |
| ```bash | |
| pip install -r requirements.txt | |
| streamlit run app.py |