Training input
- Native short-exposure RAW.
- Linear RGB after black-level normalization.
- No clean target used during training.
The same 100 reference camera poses are reused at four native short exposures. This page shows representative inputs and targets—not reconstruction outputs.























