This project is an attempt to reproduce the experiment described here [1]. Unlike previous methods directly stacking the LDR images or features for merging, this paper use attention modules to guide the merging according to the reference image. The attention modules automatically suppress undesired components caused by misalignments and saturation and enhance desirable fine details in the non-reference images. In addition to the attention model, we use dilated residual dense block (DRDB) to make full use of the hierarchical features and increase the receptive field for hallucinating the missing details. The proposed AHDRNet is a non-flow based method, which can also avoid the artifacts generated by optical-flow estimation error.
[1] Yan, Qingsen and Gong, “Attention-guided Network for Ghost-free High Dynamic Range Imaging,” The IEEE Conference on Computer Vision and Pattern Recognition (CVPR),2019.
Main code in the folder code. The training and testing datasets can be find here. “Deep high dynamic range imaging of dynamic scenes,” ACM Transactions on Graphics (Proceedings of SIGGRAPH 2017), vol. 36, no. 4, 2017.
| output | label |
|---|---|
 |
 |
 |
 |
 |
 |