25,000 images of natural scenes categorized into 6 distinct classes: buildings, forest, glacier, mountain, sea, and street. The data is organized into three directories for training, testing, and prediction to facilitate standardized model evaluation across 150x150 pixel RGB images.
Use Cases
- Train a convolutional neural network (CNN) to classify environmental scenes using the 6-class folder labels
- Evaluate model generalization on unseen data using the 7,000 unlabeled images in the 'pred' directory
- Benchmark image augmentation techniques on the 14,000 training samples to improve accuracy on the 'glacier' and 'mountain' classes
Strengths
- 25,000 total images distributed across training, testing, and prediction folders
- Six labeled categories: buildings, forest, glacier, mountain, sea, and street
- Uniform image dimensions of 150x150 pixels across the entire collection
- Pre-split architecture with 14,000 training samples and 3,000 validation samples