ConvNeXt V2-Based Deep Learning Architecture for Fine-Grained Cat Breed Classification: Implementation and Trained Models

This repository contains the complete implementation of a ConvNeXt V2-based deep learning system for fine-grained cat breed classification, achieving 88.87% validation accuracy on the CatBreedsRefined-7k dataset (20 breeds, 7,000 images). Key Features:• Pre-trained ConvNeXt V2 Base model (88M parameters) fine-tuned with progressive unfreezing strategy• Four-stage training methodology with systematic layer-wise adaptation• Comprehensive data augmentation pipeline (MixUp, CutMix, geometric/photometric transforms)• Advanced loss functions: Focal Loss (α=0.25, γ=2.0) + Label Smoothing (ε=0.1)• Mixed precision training (AMP) with gradient clipping and AdamW optimizer• Complete training logs (60 epochs), visualizations, and confusion matrix analysis Citation: If you use this implementation in your research, please cite the accompanying paper.