Reproduce “UniSER: A Foundation Model for Unified Soft Effects 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
Jingdong Zhang, Lingzhi Zhang, Qing Liu, Mang Tik Chiu, Connelly Barnes, Yizhou Wang, Haoran You, Xiaoyang Liu and 7 more
Researchers from Texas A&M University and Adobe Research developed UniSER, a foundation model that unifies the removal of diverse soft image effects like lens flare, haze, shadows, and reflections. The model achieves state-of-the-art performance on standard benchmarks and superior perceptual quality in real-world scenarios while offering fine-grained user control.
Key insights
- Diverse soft effects share the common intrinsic property of being semi-transparent occlusions, enabling a unified approach for their removal rather than separate specialist models.
- Joint-task learning across multiple soft effects is more robust and generalizable than single-task learning, leading to improved performance on individual tasks.
- High-quality, large-scale, and physically consistent synthetic data is critical for building robust foundation models for low-level vision tasks like soft effect removal.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.