PREEN predicts editor presets for exemplar color transfer
Original title: PREEN: Exemplar-Based Color Style Transfer with Smartphone Editor Presets
The model regresses 29 smartphone-editor sliders, runs at sub-second on-device latency, and leads PST50 fidelity metrics.
PREEN treats color style transfer as regression onto the 29 parameters of a deterministic image editor, outputting an interpretable preset rather than a raster. Because real style references lack preset ground truth, the authors invert a style-transfer teacher's look into pseudo-presets, pretrain a CLIP-conditioned regressor, and refine it with LoRA under a CIELAB loss. On the human-curated PST50 benchmark it leads baselines on pixel and perceptual fidelity while preserving structure. A single inference-time strength knob traces a default-to-vivid path; a two-axis study finds raters prefer vivid for matching the reference and default as a standalone photo.
Unconfirmed: Fidelity leadership and rater preferences are the authors' reported results.
- Presets are vectors of 29 editor slider values.
- Supervision uses inverted teacher outputs as pseudo-presets.
- Prediction latency is sub-second on device.
- Raters prefer the vivid end for reference match and default as a photo.
From the original article
1 min read at arxiv.orgWe present PREEN (Preset Regression via an Exemplar-based Editor Network), which casts color style transfer as regression onto the 29 parameters of a deterministic image editor. Given a content image and a style reference, PREEN predicts a preset, a vector of editor slider values that is interpretable and editor-consumable, not a raster image.
By Beomjin Ahn · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.