INTAGIUM-ZX

INTAGIUM-ZX is a specialized storage and random-access engine for segmentation masks and label maps. Instead of storing large mask datasets as many separate PNG files, INTAGIUM-ZX re-encodes them into an optimized bundle format designed specifically for AI and computer vision workflows.

The goal is simple: reduce storage size and improve random-read performance compared to standard PNG-based mask storage, while preserving the original label information. This makes INTAGIUM-ZX especially relevant for semantic segmentation pipelines, dataset handling, preprocessing, and training workflows that rely on large volumes of mask data.

INTAGIUM-ZX is not a general image compression tool. It is built specifically for segmentation masks, where fixed class schemas and efficient access matter more than generic file compression. On tested raw benchmarks, INTAGIUM-ZX showed smaller bundle sizes and faster random reads than PNG-based mask storage.

Benchmark results and technical details: See the public benchmark repository here: https://github.com/amrae1/intagium-zx-benchmarks