ARTFEED — Contemporary Art Intelligence

WildRelight: Real-World Benchmark for Single-Image Relighting

other · 2026-05-13

Researchers have introduced WildRelight, the first in-the-wild dataset for evaluating single-image relighting models. Current synthetic benchmarks fail to capture real-world complexity, causing domain shifts in state-of-the-art models. WildRelight features high-resolution outdoor scenes with aligned, temporally varying natural illuminations and high-dynamic-range environment maps. The dataset aims to bridge the synthetic-to-real gap in relighting research.

Key facts

  • WildRelight is the first in-the-wild dataset for single-image relighting evaluation.
  • Dataset includes high-resolution outdoor scenes with aligned natural illuminations.
  • Each scene is paired with a high-dynamic-range environment map.
  • State-of-the-art models trained on synthetic data show severe domain shifts.
  • Current datasets are designed for multi-view reconstruction, not relighting.
  • WildRelight addresses the synthetic-to-real gap in relighting.
  • The dataset features temporally varying natural illuminations.
  • A rigorous benchmark was established using WildRelight.

Entities

Sources