How sahilkatiyar17/deep_learning_practicals works
Project type and purpose
• This is a collection of Jupyter notebooks for deep‐learning practicals.
• Purpose: explore image classification (Cats vs Dogs), data augmentation, transfer-learning (feature extraction and fine-tuning) with and without augmentation.
File structure
• All files are notebooks (.ipynb).
• No “src” folder or standalone scripts; the code lives in the notebooks themselves.
• Key notebooks:
• Cat-V-Dogs.ipynb (base classification)
• Data_augmentation.ipynb (augmenting images)
• Transfer-learning-feature-extraction…(with and without augmentation)
• transfer_learning_finetuning…
• No explicit config or deployment files—this is a self-contained exploratory repo.
README
• Confirms the repo is for deep-learning learning exercises.
• No additional architecture details or external services specified.
a. Main components:
1. Jupyter Notebook Environment
2. Data Source (local dataset or Kaggle Cats vs Dogs)
3. Preprocessing Pipelines
– Data Loader
– Data Augmentation Module
4. Model Training Modules
– Base CNN (from scratch)
– Feature-Extraction Model (pretrained backbone)
– Fine-Tuning Model (pretrained backbone + head)
5. Evaluation & Visualization
6. Artifacts Storage (trained model weights, logs, plots)
b. Relationships & Interactions:
1. Jupyter Notebook calls Data Loader → fetch raw images
2. Raw images → Preprocessing / Augmentation → Batches
3. Batches fed into Model Training modules
4. Models output metrics & checkpoints to Artifacts Storage
5. Notebook reads checkpoints for Evaluation & Visualization
c. Architectural patterns / design principles:
• Pipeline pattern for preprocessing → training → evaluation
• Reusability: augmentation logic used across multiple notebooks
• Layered approach: separate concerns (data, model logic, evaluation)
d. Technologies / libraries:
• Python 3.x, Jupyter Notebook
• TensorFlow or PyTorch (depending on implementation)
• torchvision or keras.preprocessing.image for augmentation
• Matplotlib / Seaborn for plots
• (Optional) Kaggle API if data downloaded programmatically
• Base CNN Training
• Pretrained Feature Extraction
• Pretrained Fine-Tuning
7. Callouts listing the primary libraries used at each stage.