nhdrrnet-pytorch's Introduction

NHDRRNet-pytorch

📷 NHDRRNet (TIP'20) implementation using PyTorch framework

Introduction

This repository is the implementation of NHDRRNet [2] using PyTorch framework. The author did not open the code, therefore, we create this repository to implement NHDRRNet using PyTorch framework.

Requirements

PyTorch 1.4+
Cuda version 10.1+
OpenCV
numpy, tqdm, scipy, etc

Getting Started

Download Dataset

The Kalantari Dataset can be downloaded from https://www.robots.ox.ac.uk/~szwu/storage/hdr/kalantari_dataset.zip [2].

Dataset Model Selection

There are two dataset models provided in dataset folder. Using HDRpatches.py will generate patches in patches folder and will cost ~200GB spaces, but it runs faster. Using HDR.py (default) will open image file only when it needs to do so, thus it will save disk space. Feel free to choose the method you want.

Configs Modifications

You may modify the arguments in Configs() to satisfy your own environment, for specific arguments descriptions, see utils/configs.py.
You may modify arguments of NHDRRNet to train a better model, for specific arguments descriptions, see config dictionary in models/NHDRRNet.py.

Train

python train.py

Test

First, make sure that you have models (checkpoint.tar) under checkpoint_dir (which is defined in Configs()).

python test.py

Note. test.py will dump the result images in sample folder.

Tone-mapping (post-processing)

Generated HDR images are in .hdr format, which may not be properly displayed in your image viewer directly. You may use Photomatix for tonemapping [2]:

Download Photomatix free trial, which won't expire.
Load the generated .hdr file in Photomatix.
Adjust the parameter settings. You may refer to pre-defined styles, such as Detailed and Painterly2.
Save your final image in .tif or .jpg.

Reference

[1] Yan, Qingsen, et al. "Deep hdr imaging via a non-local network." IEEE Transactions on Image Processing 29 (2020): 4308-4322.

[2] elliottwu/DeepHDR repository: https://github.com/elliottwu/DeepHDR

nhdrrnet-pytorch's People

Contributors

Stargazers

Watchers

nhdrrnet-pytorch's Issues

Would you share your pretrained model?

Hi, @Galaxies99!

I train your model and validate the qualitative score.
However, I can not reproduce your results reported in the pepar.

So, would you share your pre-trained models?

Best,

Further speedup can be obtained by processing LDR2HDR on GPU

Hello, thank you for your sharing the code！

As i profiled your training code， i found that about 40% of the time were cost by the function LDR2HDR, which could be saved by doing this process on GPU in parallel.
A friendly tip~

Is the results of this method higher than AHDRNet?

Thanks a lots for your contribution. I look forward to obtain a baseline based on Pytorch. But I cannot re-implement the original results. I want to ask one question: is the results of this method higher than AHDRNet?

A bug in the training phase

Hello，first thanks for your code!

when I train the model follow you tips, I found that the code will go wrong when I was training. Then I found the error occurs in models/NHDRRNet.py, line 206. In more detail, the error will occur when runing the function of eval_one_epoch() after finishing one epoch.

The error info is RuntimeError: Sizes of tensors must match except in dimension 2. Got 125 and 124 (The offending index is 0), I find that the reason why the error occur is the size of the model input is unsuitable, so when they cat each other after decoder, the size will be unequal.

My solution is change 8 to 16 in utils/dataprocessor.py, line 99 & line 100. After that, the size can be equal after a serier of decoder.

My solution may not be the best, so I wish you to fix this bug, Thanks~

Recommend Projects

galaxies99 / nhdrrnet-pytorch Goto Github PK