CIFAR-100 Classification

Overview

Classify 32x32 color images into 100 fine-grained categories. CIFAR-100 is a standard benchmark for image recognition, grouped into 20 superclasses each containing 5 fine-grained classes.

Data Description

The dataset contains 60,000 color images of size 32x32 pixels:

• Training set — 50,000 images (500 per class)
• Test set — 10,000 images (100 per class)
• Fine labels — 100 classes (e.g., apple, aquarium_fish, baby, bear, beaver, ...)
• Coarse labels — 20 superclasses (e.g., fruit_and_vegetables, fish, people, ...)

Each image belongs to exactly one fine-grained class and one superclass. Images are stored as numpy arrays with shape (32, 32, 3) in RGB channel order.

Evaluation

Submissions are scored using top-1 accuracy on fine-grained labels — the fraction of test images whose predicted class matches the true class.

$$\text{Accuracy} = \frac{\text{Correct Predictions}}{\text{Total Predictions}}$$

Submission Format

Submit a CSV with exactly two columns: