yael-vinker / unpaired_hdr_tmo Goto Github PK

View Code? Open in Web Editor NEW

14.0 2.0 5.0 395.09 MB

Python 91.57% Shell 7.63% Dockerfile 0.80%

unpaired_hdr_tmo's Introduction

Unpaired Learning for High Dynamic Range Image Tone Mapping

Pytorch implementation of our method for HDR image tome mapping using GAN.

Installation

Installation via Docker [Recommended]

We provide a docker that contains the code and all the necessary libraries. It's simple to install and run.

docker build -t unpaired_tmo .
docker run --name unpaired_tmo -it -p 8888:8888 unpaired_tmo /bin/bash

Installation via Pip/Conda/Virtualenv

Clone the repo:

git clone https://github.com/yael-vinker/unpaired_hdr_tmo.git
cd unpaired_hdr_tmo

Create a new environment and install the libraries:

python3.6 -m venv hdr_venv
source hdr_venv/bin/activate
pip install -r requirements.txt

Quickstart (Run Demo Locally)

Run a model on your own image

To run the trained model of a task on a specific image:

cd activate_trained_model
python run_trained_model.py

Specify the path to a folder containing the input images using --input_images_path, the tone mapped images will be saved to the folder under --output_path.

unpaired_hdr_tmo's People

Contributors

Stargazers

Watchers

Forkers

liuguoyou killsking jackzhousz jizhe-cheer 1014456378

unpaired_hdr_tmo's Issues

About license of the code

Thank you for sharing great code!
Could you kindly let us know about the license of this project?

If we could know the exact license of this code, we can use this project for our internal R&D projects!

cuDNN error

I followed the instructions and installed the requirements in the environment.yaml but I get the following error:

Traceback (most recent call last):
  File "run_trained_model.py", line 147, in <module>
    run_trained_model(opt)
  File "run_trained_model.py", line 53, in run_trained_model
    output_images_path, args.f_factor_path, None, model_params["final_shape_addition"])
  File "run_trained_model.py", line 68, in run_model_on_path
    model_params, f_factor_path, final_shape_addition)
  File "/media/cgl/MUSKETEER_DATA/frameworks/unpaired_hdr_tmo/utils/model_save_util.py", line 209, in run_model_on_single_image
    fake = G_net(gray_im_log.unsqueeze(0), apply_crop=model_params["add_frame"], diffY=diffY, diffX=diffX)
  File "/home/cgl/miniconda3/envs/vinker/lib/python3.6/site-packages/torch/nn/modules/module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "/media/cgl/MUSKETEER_DATA/frameworks/unpaired_hdr_tmo/models/unet_multi_filters/Unet.py", line 86, in forward
    next_x = self.inc(x)
  File "/home/cgl/miniconda3/envs/vinker/lib/python3.6/site-packages/torch/nn/modules/module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "/media/cgl/MUSKETEER_DATA/frameworks/unpaired_hdr_tmo/models/unet_multi_filters/unet_parts.py", line 202, in forward
    x = self.conv(x)
  File "/home/cgl/miniconda3/envs/vinker/lib/python3.6/site-packages/torch/nn/modules/module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "/media/cgl/MUSKETEER_DATA/frameworks/unpaired_hdr_tmo/models/unet_multi_filters/unet_parts.py", line 64, in forward
    x = self.conv(x)
  File "/home/cgl/miniconda3/envs/vinker/lib/python3.6/site-packages/torch/nn/modules/module.py", line 547, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/cgl/miniconda3/envs/vinker/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 343, in forward
    return self.conv2d_forward(input, self.weight)
  File "/home/cgl/miniconda3/envs/vinker/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 340, in conv2d_forward
    self.padding, self.dilation, self.groups)
RuntimeError: cuDNN error: CUDNN_STATUS_EXECUTION_FAILED

One question about the result

Hi, I encountered one problem when I use the default belgium.hdr for inference.

There is a darker part on the upper edge of the door. I wonder if it is normal ?

Configuation for Reproduction from Scratch

Thank you for sharing your great research.

Short Quesition: What is the exact configuration (config.py) to train the model from scratch?
I would like to train your model using the training code (main_train.py) but I am somewhat unable to figure out the details of the training configuration.

Thank you.