Jigsaw Puzzle Solving is a self-supervised pretext task where the model is trained to predict the correct spatial arrangement of shuffled image patches — requiring the network to learn spatial relationships, object structure, and visual semantics.

How Does Jigsaw Puzzle Solving Work?

• Process: Divide image into a 3×3 grid (9 patches). Shuffle them into one of N predefined permutations. The network predicts which permutation was used.
• Permutations: Typically 100-1000 selected permutations (out of 9! = 362,880 total).
• Architecture: Siamese-style — each patch encoded independently, then combined for classification.
• Paper: Noroozi & Favaro (2016).

Why It Matters

• Spatial Reasoning: Forces the model to understand spatial relationships between object parts.
• Feature Quality: Learned features transfer well to object detection and segmentation tasks.
• Historical: One of the pioneering pretext tasks that launched the self-supervised learning era.

Jigsaw Puzzle Solving is teaching AI spatial common sense — learning that the head goes above the body and the legs go below, all without human labels.