Image classification using deep learning

In the framework of this project where trained two famous NN's - Inceptionv3 and EfficientNetb1. Project include following sections

• Explanatory Data Analasys
• Training pipeline

Data

• Flowers Recognition Dataset with 5 classes of flowers
• Total images: 4317 images
• Daisy: 764 images
• Dandelion: 1052 images
• Rose: 784 images
• Sunflower: 733 images
• Tulip: 984 images

Result

• EfficientNet show pretty good results with:
• Train Accuracy: 94%
• Validation Accuracy: 82%
• Precision/recall: 91%, 86%
• Matthews Corr. coef.: 0.86
• Cohen Kappa: 0.86

Used libraries and frameworks

During project where used following libraries:

• PyTorch - Deep Learning Framework
• Pandas - awesome data processing library
• torchvision - Tools for working with PyTorch CV related problems.
• Wandb - Experiment Tracker
• torchmetrics - PyTorch metrics
• sklearn - Machine Learning library

Installation

Project requires Pytorch 1.11.0.

Install the dependencies via pip or conda install managers.

conda install pytorch torchvision torchaudio cpuonly -c pytorch
conda install -c pytorch torchvision
conda install -c conda-forge torchmetrics
conda install -c conda-forge wandb