imagededup

Why we included this project

Duplicate images quietly eat storage and muddy datasets, and imagededup is built around that problem. The package includes perceptual, difference, wavelet, and average hashing for fast exact-duplicate detection, plus a CNN-based method that catches near duplicates, including images that have been resized or otherwise edited. The two tiers work well together: run the cheap hashing across a huge collection first, then bring in the neural approach where edits defeat the hash. It also comes with an evaluation framework that compares results against a ground-truth mapping, so you can verify a dedup pass actually works before wiring it into a pipeline. A few lines of Python generate encodings, find duplicates, and plot matches for visual inspection.