How anubhavparas/image-classification-using-cnn works
Project type and purpose
• This is an image‐classification pipeline built in Python using convolutional neural networks (CNNs).
• Its goal is to ingest a folder of cat/dog images, preprocess them, train a CNN, save model artifacts, and run inference producing evaluation plots and a prediction video.
File structure
• Top‐level:
• Code/ – all Python modules and folders for training & preprocessing
• README.md & Report.pdf – project description and academic report
• output/ – generated plots, MP4 prediction video
• Code/:
• constants.py – hyperparameters, filepaths
• data_prep.py – dataset loading, resizing, train/validation split
• model.py – CNN architecture definition (Keras/TensorFlow)
• main.py – orchestrates data prep, model instantiation, training loop, saving weights
• models/ – checkpointed .h5 weights and pickled history
• notebook/gpu_classification.ipynb – exploratory analysis and GPU‐enabled experimentation
README
• Describes a pipeline: data ingestion → training (15 epochs, 80/20 split) → evaluation
• Shows performance metrics, accuracy/loss graphs, sample predictions
a. Main components
• Data Store: raw image dataset (external input)
• Data Preparation Module: Code/data_prep.py + constants.py
• Model Definition Module: Code/model.py
• Training Orchestrator: Code/main.py
• Model Artifact Store: Code/models/*.h5 + *.pickle
• Evaluation & Visualization:
– Automated plots (output/*.png)
– Prediction video (output/*.mp4)
• Notebook Interface: Code/notebook/gpu_classification.ipynb for interactive runs
• Documentation: README.md + Report.pdf
b. Relationships & interactions
• Raw images → Data Preparation → formatted tensors
• Tensors + hyperparameters → Model Definition → compiled model
• Training Orchestrator passes data & model → GPU/CPU compute → training loop
• At each checkpoint → save weights to Model Artifact Store
• Post‐training → load weights → run evaluation → generate plots & video in output/
• Notebook can tap into Data Preparation, Model Definition, and Artifact Store for interactive experiments
c. Architectural patterns/design principles
• Pipeline pattern: clear separation of data preparation, model definition, training, and evaluation
• Modularity: each concern isolated in its own file
• Reproducibility: constants.py centralizes hyperparameters, model artifacts are versioned
d. Key technologies & libraries
• Python 3.x
• Keras/TensorFlow (model serialization to .h5)
• pickle (history serialization)
• Jupyter Notebook
• Matplotlib/seaborn (plots)
• FFmpeg or OpenCV (video generation)