Reproduce “DehazeNet: An End-to-End System for Single Image Haze Removal” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bolun Cai, Xiangmin Xu, Kui Jia, Chunmei Qing, Dacheng Tao
DehazeNet introduces an end-to-end Convolutional Neural Network for single image haze removal, learning to estimate the medium transmission map directly from hazy inputs. It demonstrates superior accuracy, robustness to varying haze conditions and noise, and real-time processing capability compared to traditional prior-based and early learning-based methods.
Key insights
- Maxout units can effectively learn and emulate extremum operations for haze-relevant feature extraction, improving convergence over standard activations.
- BReLU, a novel bounded activation function, significantly improves the accuracy and physical plausibility of estimated transmission maps by constraining outputs within physically valid ranges.
- Integrating multi-scale feature extraction in CNNs enhances robustness to varying object scales and haze conditions in image restoration tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.