Algorithmic reduction of data volume — common in downsampling high-res footage or polygon reduction in 3D. Fast but introduces artifacts.
You encounter decimation daily in the VFX pipeline—though mostly unconsciously. As soon as you work with high-resolution raw data, whether it's 8K footage or millions of polygons from a 3D scan, you'll need to decimate at some point. This means systematically discarding data to reduce file size and processing time. It's not just scaling—decimating means taking every n-th value or interpolating more intelligently.
With image sequences, it works very simply: You have an 8K ProRes Raw project, but you need your proxies in 4K or HD. A simple decimation algorithm takes every second (or fourth) pixel and discards the rest. Fast, but crude—edges become jagged, fine details disappear. In the motion graphics context, when you reduce 120 fps footage to 24 fps, this is often perfectly sufficient. A worse problem: you lose high-frequency image information that can become critical later in color grading. That's why experienced colorists work with anti-aliased downsampling instead of pure decimation—filtering is applied beforehand to avoid aliasing artifacts.
In the 3D pipeline, decimation is your friend when scan data or CAD models arrive with millions of polygons. A decimation plugin (like in ZBrush, Maya, or Houdini) reduces polygon density while preserving silhouettes and details as much as possible. This is not the same as subdivision or retopology—decimation retains the original topology, just throws away polygons. In a high-poly to low-poly workflow, you save enormous render times but pay with a loss of detail. You notice this particularly during extreme camera moves, when the decimated geometry feels unnaturally angular.
Practically speaking: Decimation is a last resort, not a creative tool. It buys you time and storage, but always costs you visual quality. Unlike more intelligent compression methods (such as adaptive sampling or wavelet compression), decimation is brutal and predictable—ideal for fast proxies, but not for final output. In modern VFX pipelines, you often need both: decimated working versions for quick iteration, and high-fidelity raw data for final rendering.