This project aims to classify natural scenes into six categories by applying Convolutional Neural Network (CNN), transfer learning, and experiment tracking. The dataset used is the Intel Image Classification dataset, which includes 25k+ images of various types of landscapes and environments.
-
Data Preprocessing: Includes resizing, normalization, and data augmentation to improve model performance.
-
CNN Architecture: Applies transfering on the pretrained model, ResNet18.
-
Experiment Tracking: Using TensorBoard to track training loss and accuracy.
The dataset is sourced from Intel Image Classification Dataset. It contains the following categories:
- Buildings
- Forest
- Glacier
- Mountain
- Sea
- Street
-
Accuracy: Achieved an accuracy of 90% on the test set after 5 training epochs (Only trained for 5 epochs due to hardware limitations).
-
Loss and accuracy plot:
