By Zhen Liu^1,2, Wenjie Lin¹, Xinpeng Li¹, Qing Rao¹, Ting Jiang¹, Mingyan Han¹, Haoqiang Fan¹, Jian Sun¹ and Shuaicheng Liu^3,1*

¹Megvii Technology, ²Sichuan University, ³University of Electronic Science and Technology of China

This is the official Pytorch implementation of ADNet: Attention-guided Deformable Convolutional Network for High Dynamic Range Imaging.

arXiv

In this paper, we present an attention-guided deformable convolutional network for hand-held multi-frame high dynamic range (HDR) imaging, namely ADNet. This problem comprises two intractable challenges of how to handle saturation and noise properly and how to tackle misalignments caused by object motion or camera jittering. To address the former, we adopt a spatial attention module to adaptively select the most appropriate regions of various exposure low dynamic range (LDR) images for fusion. For the latter one, we propose to align the gamma-corrected images in the feature-level with a Pyramid, Cascading and Deformable (PCD) alignment module. The proposed ADNet shows state-of-the-art performance compared with previous methods, achieving a PSNR-l of 39.4471 and a PSNR-µ of 37.6359 in NTIRE 2021 Multi-Frame HDR Challenge.

• Python 3.7.0
• Pytorch 1.2.0
• torchvision 0.4.0
• CUDA 10.0 on Ubuntu 18.04

1. Install python environment

   conda create -n adnet python=3.7
   conda activate adnet
   pip install -r requirement.txt
   

2. Build DCN for PCD module

cd DCN_v2
./make.sh    # build dcn
python testcuda.py # run examples and gradient check on gpu

Download the dataset from NTIRE2021 HDR Challenge. To training our model, we first crop the training images to size 256x256 patches with a stride of 128.

python train.py

Part of our codes are adapted from AHDRNet and EDVR we thank the authors for their contributions.

@inproceedings{liu2021adnet,
  title={ADNet: Attention-guided deformable convolutional network for high dynamic range imaging},
  author={Liu, Zhen and Lin, Wenjie and Li, Xinpeng and Rao, Qing and Jiang, Ting and Han, Mingyan and Fan, Haoqiang and Sun, Jian and Liu, Shuaicheng},
  booktitle={CVPRW},
  pages={463--470},
  year={2021}
}