DAN: Unfolding the Alternating Optimization for Blind Super Resolution

Last update: Dec 13, 2022

Related tags

Deep Learning DAN-Basd-on-Openmmlab

Overview

DAN-Basd-on-Openmmlab

DAN: Unfolding the Alternating Optimization for Blind Super Resolution

We reproduce DAN via mmediting based on open-sourced code.

Requirements

PyTorch >= 1.3
mmediting >= 0.9

DataSets

We use DIV2K and Flickr2K as our training datasets. For evaluation of Setting 2, we use DIV2KRK datasets,

Usages

How to run this repo: copy the file to the mmediting workspace and run the program directly based on the commands in mmediting

Copy files to MMEditing workspace.

cd DAN-Basd-on-Openmmlab/
mv ./mmedit/models/restorers/dan.py ${mmediting_workspace}/mmedit/models/restorers/
mv ./mmedit/models/backbones/sr_backbones/dan_net.py ${mmediting_workspace}/mmedit/models/backbones/sr_backbones/
mv ./mmedit/models/common/DANpreprocess.py ${mmediting_workspace}/mmedit/models/common
mv ./configs/restorers/dan ${mmediting_workspace}/configs/restorers/
mv ./tools/data/super-resolution/dan_datasets ${mmediting_workspace}/tools/data/super-resolution/

Modify the configuration file as follows:

pca_matrix_path='${mmediting_workspace}/tools/data/super-resolution/div2k/pca_matrix/pca_aniso_matrix_x4.pth' # your pca_matrix path
# Training
gt_folder='${dataset_workspace}/dataset/DF2K_train_HR_sub' # your train data path
# Testing
lq_folder='${dataset_workspace}/dataset/DIV2KRK/lr_x4' # your test data LR path
gt_folder='${dataset_workspace}/dataset/DIV2KRK/gt' # your test data HR path

Add script to init file, as follows:

modify the mmedit/models/backbones/sr_backbones/__init__.py:

from .dan_net import DAN
# add DAN into __all__ list.

modify the mmedit/models/commons/__init__.py:

from .dan_preprocess import SRMDPreprocessing
# add SRMDreprocessing into __all__ list.

modify the mmedit/models/restorers/__init__.py:

from .dan import DAN
# add DAN into __all__ list.

Training/Test

Before using it, please download and process the dataset and set the path in the configuration file.

Train

# Single GPU
python tools/train.py configs/restorers/dan/dan_setting2.py --work_dir ${YOUR_WORK_DIR}

# Multiple GPUs
./tools/dist_train.sh configs/restorers/dan/dan_setting2.py ${GPU_NUM} --work_dir ${YOUR_WORK_DIR}

Test

# Single GPU
python tools/test.py configs/restorers/dan/dan_setting2.py ${CHECKPOINT_FILE} [--metrics ${METRICS}] [--out ${RESULT_FILE}]

# Multiple GPUs
./tools/dist_test.sh configs/restorers/dan/dan_setting2.py ${CHECKPOINT_FILE} ${GPU_NUM} [--metrics ${METRICS}] [--out ${RESULT_FILE}]

Result

DIV2KRK

The passwds of download links are all 'ta2o'

Method	scale	Datasets	PSNR	Download
DAN	x4	DIV2KRK	27.41	model / test_pkl

DAN: Unfolding the Alternating Optimization for Blind Super Resolution

Related tags

Overview

DAN-Basd-on-Openmmlab

Requirements

DataSets

Usages

Result

DIV2KRK

Owner

AlexZou

The mini-MusicNet dataset

nnFormer: Interleaved Transformer for Volumetric Segmentation

DANet for Tabular data classification/ regression.

A multi-entity Transformer for multi-agent spatiotemporal modeling.

Real-Time Multi-Contact Model Predictive Control via ADMM

The fastai deep learning library

Deep Inside Convolutional Networks - This is a caffe implementation to visualize the learnt model

Convert dog pictures into various painting styles. Try LimnPet

CrossNorm and SelfNorm for Generalization under Distribution Shifts (ICCV 2021)

Deep High-Resolution Representation Learning for Human Pose Estimation

Using a Seq2Seq RNN architecture via TensorFlow to predict future Bitcoin prices

TorchOk - The toolkit for fast Deep Learning experiments in Computer Vision

An official reimplementation of the method described in the INTERSPEECH 2021 paper - Speech Resynthesis from Discrete Disentangled Self-Supervised Representations.

Code for the ICCV 2021 Workshop paper: A Unified Efficient Pyramid Transformer for Semantic Segmentation.

An alarm clock coded in Python 3 with Tkinter

YOLOX-CondInst - Implement CondInst which is a instances segmentation method on YOLOX

Code for our paper 'Generalized Category Discovery'

Synthetic LiDAR sequential point cloud dataset with point-wise annotations

Fast Style Transfer in TensorFlow

Experiments with differentiable stacks and queues in PyTorch