Open tkuri opened 3 years ago
RAW画像からのデモザイクとノイズ除去を行う。Dataset中のGround Truthも画像処理プロセス(ISP)を通っているので劣化しているのでそのまま学習するのは問題であると指摘し真のGTからの劣化過程をモデル化し真のGTらしさ(尤度)を最大化するようにLossを定式化したのがキモ。
https://arxiv.org/abs/2101.04442
未確認。
論文概要
RAW画像からのデモザイクとノイズ除去を行う。Dataset中のGround Truthも画像処理プロセス(ISP)を通っているので劣化しているのでそのまま学習するのは問題であると指摘し真のGTからの劣化過程をモデル化し真のGTらしさ(尤度)を最大化するようにLossを定式化したのがキモ。
https://arxiv.org/abs/2101.04442
Code
未確認。