A plug-and-play loss that trains RAW reconstruction networks to remain accurate after diverse photofinishing edits by comparing rendered sRGB outputs from a randomly parameterized differentiable ISP.
Beyond learned metadata-based raw image reconstruction.International Journal of Computer Vision, 132(12):5514–5533
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Edit-aware RAW Reconstruction
A plug-and-play loss that trains RAW reconstruction networks to remain accurate after diverse photofinishing edits by comparing rendered sRGB outputs from a randomly parameterized differentiable ISP.