Introducing UWFormer, a Multi-Scale Transformer for enhancing underwater images through semi-supervised learning.
提案されたUWFormerは、多頻度の画像をセミスーパーバイズド学習を用いて強化する革新的なマルチスケールトランスフォーマーベースのネットワークです。
The core message of this paper is to present an improved Cycle GAN based model for underwater image enhancement that utilizes depth-oriented attention to enhance the contrast of the overall image while keeping global content, color, local texture, and style information intact.
MambaUIE, a novel state-space modeling-based architecture, efficiently enhances underwater images by capturing global contextual information and local fine-grained features.
MambaUIE는 효율적인 상태 공간 모델 기반 아키텍처를 통해 매우 적은 연산량으로도 높은 정확도의 수중 이미지 향상을 달성한다.
A novel depth-guided perception framework, UVZ, is proposed to effectively enhance the color, contrast, and clarity of underwater images by adaptively combining non-local and local features.
The proposed FDCE-Net effectively enhances underwater images by decoupling degradation factors in the frequency domain and learning semantic-aware color representations, resulting in improved color fidelity, texture details, and overall visual quality compared to state-of-the-art methods.
A physics-aware deep learning network that explicitly estimates the degradation parameters of the underwater image formation model, combined with an IFM-inspired semi-supervised learning framework, to effectively enhance underwater images while addressing the challenge of insufficient labeled data.
A novel end-to-end network called WaterFormer is proposed to enhance underwater images by effectively addressing the independent yet interdependent issues of haze and color degradation.
The proposed PDCFNet network leverages pixel difference convolution and cross-level feature fusion to effectively enhance the visibility, detail, and color of underwater images.