CIFAR-10 class distribution
View the interactive bar chart of the training set class distribution, rendered with Plotly in dark mode.
Open plot ↗Examples per class
Explore a grid of example images for each CIFAR-10 class to build an intuitive understanding of the dataset.
Open plot ↗Accuracy over epochs
Inspect how training and validation accuracy evolve over time to check convergence behaviour and potential overfitting.
Open plot ↗Loss over epochs
Follow the training and validation loss curves to understand optimisation progress and stability during training.
Open plot ↗Class-wise confusion
Explore the normalised confusion matrix to see which classes are reliably recognised and which ones are frequently confused.
Open plot ↗Per-class accuracy
Compare accuracy across all CIFAR-10 classes to identify particularly strong and weak categories of the model.
Open plot ↗Average confidence per true class
See how confident the model is on average for the true label of each class, even when it makes mistakes.
Open plot ↗Confidence: correct vs wrong
Compare the confidence distribution of correct and wrong predictions to understand how well-calibrated the model is.
Open plot ↗Top 1% most confident correct
Browse the samples where the model is extremely confident and correct – these represent the easiest cases for the network.
Open plot ↗Top 1% most confident wrong
Inspect the most confidently wrong predictions to reveal surprising failure modes and visually similar confusing examples.
Open plot ↗Hard but correct predictions
View samples that are correctly classified but with low confidence. These are borderline cases where the model is uncertain.
Open plot ↗Misclassification grid
Explore a grid of misclassified images to visually inspect how the predicted class differs from the true label.
Open plot ↗