This repository is the tensorflow implementation for the HDR deghosting, AHDR-Net, proposed in the paper - Attention-guided Network for Ghost-free High Dynamic Range Imaging.

system requirements: Python:3.5.2
numpy==1.16.4
tensorboard==1.13.1
tensorflow-estimator==1.13.0
tensorflow-gpu==1.13.1
opencv-python==4.1.0.25

The training script for this model is fusion_train_256_ahdr.py.The model was trained on UCSD dataset for 50 epochs with batch size of 8.

`python ahdr_test.py

Test outputs consist of predicted hdr images and tonemapped its version.

For further queries, please mail at durgesh080793 <at> gmail <dot> com.