Dog Breed Classifier

Fine-grained image classification with ResNet50V2 transfer learning.

Explore the notebook · Project report · Execution report

Overview

This notebook builds an end-to-end dog-breed classification pipeline on the Kaggle Dog Breed Identification dataset. It prepares image and label data, augments the training set, uses an ImageNet-pretrained ResNet50V2 backbone, trains a custom classification head, and runs inference on unseen dog images.

Experiment	Configuration
Dataset	Kaggle Dog Breed Identification — 120 source labels
Training scope	60 selected breed classes
Input	224 × 224 RGB images
Model	Frozen ResNet50V2 backbone + batch normalization + dense head
Split	80% training / 20% validation
Training	RMSprop, batch size 64, 20 epochs
Best recorded validation accuracy	80.66%

Prediction samples

Pipeline

Kaggle images + labels
        ↓
Resize and ResNet preprocessing
        ↓
Train/validation split + augmentation
        ↓
ResNet50V2 feature extractor
        ↓
Classification head and training
        ↓
Breed prediction for a new image

Run the notebook

1. Clone the repository and create a Python environment.

2. Install the dependencies:

   pip install -r requirements.txt

3. Configure the Kaggle API locally. Never commit kaggle.json.

4. Open dog_breed_detection.ipynb in Jupyter or Google Colab and run the cells in order.

The trained model and downloaded dataset are intentionally excluded from version control because they are generated artifacts.

Repository layout

.
├── dog_breed_detection.ipynb    # data preparation, training and inference
├── test images/                  # small inputs used for inference checks
├── docs/assets/                  # README visuals and clean prediction crops
├── requirements.txt
└── README.md

---

Built by Rahul Tripathi · Portfolio · LinkedIn