Restore Face (GFPGAN)

How it fits

Typical backends

Caveats

• I/O contractThe output is a downstream-resolved oneof t[Image, UInt64, [Image]] of the same restoration: the full restored frame, the UInt64 face count, or the list of per-face 512x512 crops. The count equals the crops length for every frame. Output pixels are RGB regardless of the input colour order.
• FallbackWhen no face is detected the Image arm returns the source enlarged by upscale, the UInt64 arm 0, and the [Image] arm an empty list; a zero-sized frame is passed through on the full-frame arm.
• Hard constraintThe configured weights file must match the architecture chosen by arch; a mismatched pair loads incorrectly. Both are captured once at startup, so switching variants requires a redeploy.
• Parameter interactionupscale enlarges the whole output frame by that factor on top of face restoration, so a larger factor raises memory and time roughly with the output area; the per-face crops stay 512x512 regardless.
• CompatibilityGFPGAN downloads its own face detection and parsing weights from the public release host on first use, so a cold container on a restricted network must be able to reach them or pre-seed the cache.
• CostA GPU is strongly preferred; CPU face restoration is slow for anything beyond occasional single images.