This repository is the pytorch implementation of various High Dynamic Range (HDR) Imaging algorithms. Please find the details below.

@TianhongDai and @WeiLi-THU

• pytorch==1.4.0
• opencv-python
• scikit-image==0.17.2

• adaptive padding
• add more baselines

• DeepHDR [1]
• NHDRRNet [2]
• AHDR [3]
• DAHDR [4]

1. download the Kalantari dataset via: [link], and organize the dataset as follows:

dataset
│
└───Traning
│   │  001
│   │  002
│   │  003
│   |  ...
│   
└───Test
    │  001
    │  002
    |  003
    |  ...   

2. train the network [unet|nhdrrnet|ahdr|dahdr]:

python train.py --net-type unet --cuda --batch-size 8 --lr 0.0002

3. continue training using the pre-saved checkpoint:

python train.py --net-type unet --cuda --resume --last-ckpt-path <the saved ckpt path> 

4. test the model and save HDR images:

python eval_metric.py --net-type unet --model-path <the saved ckpt path> --cuda --save-image

The pre-trained models can be downloaded from the released page.

@elliottwu for DeepHDR
@qingsenyangit for AHDRNet
@Galaxies99 for NHDRRNet details

[1] Deep High Dynamic Range Imaging with Large Foreground Motions
[2] Deep HDR Imaging via A Non-Local Network
[3] Attention-guided Network for Ghost-free High Dynamic Range Imaging
[4] Dual-Attention-Guided Network for Ghost-Free High Dynamic Range Imaging