CMA: A Chromaticity Map Adapter for Robust Detection of Screen-Recapture Document Images

Overview

This is the implementation of the method proposed in "CMA: A Chromaticity Map Adapter for Robust Detection of Screen-Recapture Document Images" with pytorch(1.7.1, gpu version).

Introduction

In this work, we design the Chromaticity Map Adapter (CMA) network to extract chromaticity feature maps based on the pixel-level distortion model established in the paper. These feature maps are then fed into the Transformer network as multi-modal prompt tokens.

Reference

This work has been accepted by CVPR2024. The code and data will be available soon.

[1] C. Chen, L. Lin, Y. Chen, B. Li, J. Zeng, J. Huang, "CMA: A Chromaticity Map Adapter for Robust Detection of Screen-Recapture Document Images," IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Accepted Feb. 2024.