Difference
Highlight every pixel where two model inferences agree or disagree.
Original introduction
See what the product was built to do
The original walkthrough remains the clearest introduction to the problem and the workflow.
GoodModelBadModel
Next step in Data Visualization
Transforming the way machine-learning teams interact with and understand complex model output.
The original product
Move from aggregate scores to visual evidence
GoodModelBadModel brought model output, ground truth, comparisons, and error analysis into one visual workflow.
Improvement
See exactly where a newer model improved, regressed, or changed without becoming more accurate.
Confusion Matrix
Understand which semantic classes a model confuses within an individual image.
Model Confusion Matrix
Aggregate the evidence across a dataset and compare model behavior at a glance.
Data
Datasets
The live product managed images, ground truth, and multiple model runs as reusable datasets.
Interactive archive
Put two models under the microscope
Pick a frame, switch views, and inspect the visual evidence. All computation happens locally in your browser—there is no account or backend.
Loading model output…
Current frame
Improved-model confusion matrix
Rows are ground-truth classes; columns are predictions. Darker cells contain more sampled pixels.
Preserved product experiment
Good ideas are still worth showing.
The commercial service is retired. This archive preserves the problem, the interaction model, and a working slice of the product as part of Andriy Drozdyuk's portfolio.