# FlowSteer100

Benchmark dataset for zero-shot image restoration with [FlowSteer](https://arxiv.org/abs/2512.08125) (CVPR 2026 - Findings track).
Interactive and visual comparisons are in the [Project page](https://tharindu-nirmal.github.io/FlowSteer/)

Contains 100 ground truth images for four tasks:
- Super-resolution (4×)
- Deblurring
- Denoising
- Colorization

## Usage
Used to evaluate FlowSteer in the paper.
To simulate the forward models that simulate the degradations, and the code for FlowSteer, see the [GitHub repo](https://github.com/Tharindu-Nirmal/FlowSteer) .

## License
CC BY 4.0


---
## task_categories:
- image-to-image
## tags:
- imagerestoration
- superresolution
- colorization
- deblurring
- denoising
- flux
- zeroshot
pretty_name: FlowSteer100
size_categories:
- n<1K
---