Deep Learning
Image Classifier (Cat vs Dog)
Deep learning image classifier achieving 96-98% accuracy using CNN and MobileNetV2 on 10,000 images.
Project Overview
Binary image classification comparing Custom CNN from scratch vs MobileNetV2 Transfer Learning on 10,000 real images. Shows the power of transfer learning over training from scratch.
⚠ The Problem
Pet technology platforms and content moderation systems need accurate image classification to automatically categorize uploaded photos.
✓ The Solution
Built two models: Custom CNN with 3 conv blocks and MobileNetV2 with fine-tuned last 30 layers. Data augmentation pipeline for robust generalization.
Results & Outcomes
MobileNetV2 achieves 96-98% test accuracy. ROC curve and confusion matrix comparison shows significant transfer learning advantage.