TrivialAugment is an extremely simple augmentation strategy that applies a single randomly selected augmentation with a random magnitude to each image — with zero hyperparameters, yet matching or outperforming RandAugment and AutoAugment.

How Does TrivialAugment Work?

• Sample: Pick one augmentation uniformly at random from the pool.
• Magnitude: Sample a random magnitude uniformly from the valid range.
• Apply: Apply the single augmentation to the image.
• That's It: No $N$, no $M$, no policy, no search. Zero hyperparameters.
• Paper: Müller & Hutter (2021).

Why It Matters

• Zero Hyperparameters: The simplest possible automated augmentation — no tuning at all.
• Competitive: Matches or exceeds RandAugment and AutoAugment on ImageNet, CIFAR-10, CIFAR-100.
• Lesson: Over-engineering augmentation policies may not be necessary — randomness works.

TrivialAugment is the laziest augmentation strategy that works — randomly applying one augmentation at random strength, yet matching sophisticated learned policies.