Planet SuperRes

Planet SuperRes is Planet's generative-AI enhancement for PlanetScope satellite imagery. PlanetScope gives near-daily global coverage at 3 m pixels, which leaves small objects and textures hard to make out; SuperRes sharpens those pixels into a 2 m resolution prediction designed for human visual analysis, making features such as individual trees and rooftop markings visible.

The model is an Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) trained with a perceptual loss on more than 120,000 pairs of 3 m PlanetScope and 0.5 m SkySat scenes of the same locations, each pair captured within 12 hours. SkySat is used only for training: the output is generated from PlanetScope data alone, so SuperRes is available wherever PlanetScope imagery exists, including the SuperDove archive from 2020. Every output carries a per-pixel confidence layer (0 to 100) that flags where the model has inferred detail rather than resolved it, and Planet states that AI-enhanced data should be validated by a human before use.

SuperRes comes in two forms. SuperRes PlanetScope Scenes are applied on demand to PlanetScope data in a Planet Insights Platform collection, through the browser-based Visual Analysis tool or the Processing and Batch Processing APIs. SuperRes Mosaics are visual mosaics of broad areas at weekly, biweekly, monthly or quarterly intervals. Use is metered in Planet monitoring credits.