Intel Image Classification

Scene classification on the Intel Image Classification dataset using ResNet50 transfer learning — 92.5% test accuracy across 6 natural scene categories.

---

Problem

Classify outdoor scene images into 6 categories: buildings, forest, glacier, mountain, sea, street — a 6-class multi-label image classification task.

Dataset

Intel Image Classification — 17,034 images split across train/test/prediction sets.

Split	Images
Train	14,034
Test	3,000
Prediction	7,301

Approach

• Model: ResNet50 pretrained on ImageNet, final FC layer replaced for 6-class output
• Optimizer: Adam (lr=0.0001)
• Epochs: 10
• Augmentation: Random horizontal flip, random rotation (10°)
• Framework: PyTorch + torchvision

Results

Metric	Score
Test Accuracy	92.5%
Weighted Precision	0.926
Weighted Recall	0.925
Weighted F1	0.925

Training accuracy reached 97.4% by epoch 10 with consistent loss reduction from 0.275 → 0.071.

Repo Structure

intel-image-classification/
├── notebooks/
│   └── intel_image_classification_resnet50.ipynb
├── reports/
│   └── images/          # confusion matrix, training curves
└── README.md

Setup

git clone https://github.com/VenuBandi/intel-image-classification
cd intel-image-classification
pip install -r requirements.txt

Download the dataset from Kaggle and place it as intel_dataset/ in the project root.

Related Work

• histopathologic-cancer-detection — medical imaging CNN
• structural-crack-detection — EfficientNet on SDNET2018