How leminhhaii/asl_classification_project works
Project type and purpose
• This is an end-to-end machine-learning pipeline for American Sign Language recognition.
• It provides utilities to collect or import a dataset, train a convolutional neural network (FCNN) via Jupyter notebooks, and run inference in real time on video or on static images using pre-trained model weights.
File structure
• Top-level folders:
• create dataset – scripts for capturing images and assembling the training dataset
• training – Jupyter notebooks for model development and training
• model using – Python scripts for inference (video & picture)
• example output – sample output video with predictions
• Configuration/build: requirements.txt defines dependencies; .vscode contains editor settings; .gitattributes for repo config.
• No web front end or database; purely a data-science pipeline and inference application.
README for additional insights
• Describes how to download a pre-trained model and dataset from Google Drive.
• Instructs how to run inference scripts from the project root.
• Mentions no server or UI beyond console/terminal output and optional saved media.
Main components and interactions
a. Data Layer
• Data Capture: create dataset/collect_imgs.py
• Dataset Assembly: create dataset/create_dataset.py
• External Storage: Downloaded dataset (dataset folder)
b. Training Layer
• Jupyter Notebooks (training/*.ipynb)
• ML Framework (TensorFlow or PyTorch; see requirements.txt)
• Output: Serialized model weights (saved to model folder)
c. Model Storage
• Local “model” directory containing .h5/.pth files
• External Model Source: Google Drive link for downloading pre-trained model
d. Inference Layer
• Video Inference: model using/model_use_video_camera.py
• Image Inference: model using/model_use_picture.py
• Input Sources: Webcam or image file paths
• Output: On-screen labels and saved video/image predictions (example output)
e. Dependencies & Environment
• requirements.txt (OpenCV, NumPy, ML libraries, etc.)
• Python interpreter, optional virtual environment
Architectural patterns and design principles
• Clear separation of concerns:
1) Data ingestion & preprocessing
2) Model development & training
3) Model storage & versioning
4) Inference application
• Pipeline pattern (ETL → Train → Deploy → Inference)
• Reproducibility: notebooks document experiments, requirements.txt fixes dependencies
Final recommendations
• Readability: consistent font, spacing, and alignment.