This is the official implementation of the solution proposed by team Eureka for the AIM 2020 Challenge on Learned Image Signal Processor Pipeline at ECCV 2020.
- The detailed solution can be found here.
First row: RAW images, second row: RGB images obtained by the proposed method, third row: Ground truth images.
The code was written with Python 3.6.8 with the following dependencies:
- cuda release 9.0, V9.0.176
- tensorflow 1.12.0
- keras 2.2.4
- numpy 1.16.4
- scipy 1.2.1
- imageio 2.5.0
- skimage 0.15.0
- matplotlib 3.1.0
- cuDNN 7.4.1
This code has been tested in Ubuntu 16.04.6 LTS with 4 NVIDIA GeForce GTX 1080 Ti GPUs (each with 11 GB RAM).
git clone https://github.com/puneesh00/deep_isp.git
python main.py -exp isp -dataset (full path to dataset directory) -save (full path to the repository)
There are additional parameters --resume_weight (name of weight file to resume training from), --resume_train (True if resuming training from some previous epoch), --resume_opt (name of optimizer file to resume training from). Other parameters like epochs, lr, can be changed as required.
Download weights for the model, and place them in the cloned git repository. They can be found here.
python infer_full.py -path (give full path to the repository) -w weights2_0191.h5 -dataset (path to full resolution raw images)
This will generate the output images in a folder results
(default name) in the git repository.
python infer.py -path (give full path to the repository) -w weights2_0191.h5 -dataset (path to cropped raw images) -res results_cropped
This will generate the output images in a folder results_cropped
in the git repository.
Copyright 2020 Authors
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.