Metadata-Version: 2.1
Name: VSR
Version: 1.0.4
Summary: Video Super-Resolution Framework
Home-page: https://github.com/LoSealL/VideoSuperResolution
Author: Wenyi Tang
Author-email: wenyitang@outlook.com
License: MIT
Description: # Video Super Resolution
        A collection of state-of-the-art video or single-image super-resolution architectures, reimplemented in tensorflow.
        
        **Project uploaded to PyPI now. Try install from PyPI:**
        ```shell script
        pip install VSR
        ```
        
        **Pretrained weights is uploading now.**
        
        **Several referenced PyTorch implementations are also included now.**
        
        **Quick Link:**
        - [Installation](#install)
        - [Getting Started](#Getting-Started)
        - [Benchmark](https://github.com/LoSealL/VideoSuperResolution/blob/master/Docs/Benchmark%20(reproduce).md)
        
        ## Network list and reference (Updating)
        The hyperlink directs to paper site, follows the official codes if the authors open sources.
        
        All these models are implemented in **ONE** framework.
        
        |Model |Published |Code* |VSR (TF)**|VSR (Torch)|Keywords|Pretrained|
        |:-----|:---------|:-----|:---------|:----------|:-------|:---------|
        |SRCNN|[ECCV14](https://arxiv.org/abs/1501.00092)|-, [Keras](https://github.com/qobilidop/srcnn)|Y|Y| Kaiming |[√](https://github.com/LoSealL/Model/releases)|
        |RAISR|[arXiv](https://arxiv.org/abs/1606.01299)|-|-|-| Google, Pixel 3 ||
        |ESPCN|[CVPR16](https://arxiv.org/abs/1609.05158)|-, [Keras](https://github.com/qobilidop/srcnn)|Y|Y| Real time |[√](https://github.com/LoSealL/Model/releases)|
        |VDSR|[CVPR16](https://arxiv.org/abs/1511.04587)|-|Y|Y| Deep, Residual |[√](https://drive.google.com/open?id=1hW5YDxXpmjO2IfAy8f29O7yf1M3fPIg1)|
        |DRCN|[CVPR16](https://arxiv.org/abs/1511.04491)|-|Y|Y| Recurrent ||
        |DRRN|[CVPR17](http://cvlab.cse.msu.edu/pdfs/Tai_Yang_Liu_CVPR2017.pdf)|[Caffe](https://github.com/tyshiwo/DRRN_CVPR17), [PyTorch](https://github.com/jt827859032/DRRN-pytorch)|Y|Y| Recurrent ||
        |LapSRN|[CVPR17](http://vllab.ucmerced.edu/wlai24/LapSRN/)|[Matlab](https://github.com/phoenix104104/LapSRN)|Y|-| Huber loss ||
        |EDSR|[CVPR17](https://arxiv.org/abs/1707.02921)|-|Y|Y| NTIRE17 Champion |[√](https://github.com/LoSealL/Model/releases)|
        |SRGAN|[CVPR17](https://arxiv.org/abs/1609.04802)|-|Y|-| 1st proposed GAN ||
        |VESPCN|[CVPR17](https://arxiv.org/abs/1611.05250)|-|Y|Y| VideoSR |[√](https://drive.google.com/open?id=19u4YpsyThxW5dv4fhpMj7c5gZeEDKthm)|
        |MemNet|[ICCV17](https://arxiv.org/abs/1708.02209)|[Caffe](https://github.com/tyshiwo/MemNet)|Y|-|||
        |SRDenseNet|[ICCV17](http://openaccess.thecvf.com/content_ICCV_2017/papers/Tong_Image_Super-Resolution_Using_ICCV_2017_paper.pdf)|-, [PyTorch](https://github.com/wxywhu/SRDenseNet-pytorch)|Y|-| Dense |[√](https://drive.google.com/open?id=1aXAfRqZieY6mTfZUnErG84-9NfkQSeDw)|
        |SPMC|[ICCV17](https://arxiv.org/abs/1704.02738)|[Tensorflow](https://github.com/jiangsutx/SPMC_VideoSR)|T|Y| VideoSR ||
        |DnCNN|[TIP17](http://ieeexplore.ieee.org/document/7839189/)|[Matlab](https://github.com/cszn/DnCNN)|Y|Y| Denoise |[√](https://github.com/LoSealL/Model/releases)|
        |DCSCN|[arXiv](https://arxiv.org/abs/1707.05425)|[Tensorflow](https://github.com/jiny2001/dcscn-super-resolution)|Y|-|||
        |IDN|[CVPR18](https://arxiv.org/abs/1803.09454)|[Caffe](https://github.com/Zheng222/IDN-Caffe)|Y|-| Fast |[√](https://drive.google.com/open?id=1Fh3rtvrKKLAK27r518T1M_JET_LWZAFQ)|
        |RDN|[CVPR18](https://arxiv.org/abs/1802.08797)|[Torch](https://github.com/yulunzhang/RDN)|Y|-| Deep, BI-BD-DN ||
        |SRMD|[CVPR18](https://arxiv.org/abs/1712.06116)|[Matlab](https://github.com/cszn/SRMD)|-|Y| Denoise/Deblur/SR |[√](https://drive.google.com/open?id=1ORKH05-aLSbQaWB4qQulIm2INoRufuD_)|
        |DBPN|[CVPR18](https://arxiv.org/abs/1803.02735)|[PyTorch](https://github.com/alterzero/DBPN-Pytorch)|Y|Y| NTIRE18 Champion |[√](https://drive.google.com/open?id=1ymtlOjhkGmad-od0zw7yTf17nWD4KMVi)|
        |ZSSR|[CVPR18](http://www.wisdom.weizmann.ac.il/~vision/zssr/)|[Tensorflow](https://github.com/assafshocher/ZSSR)|-|-| Zero-shot ||
        |FRVSR|[CVPR18](https://arxiv.org/abs/1801.04590)|[PDF](https://github.com/msmsajjadi/FRVSR)|T|Y| VideoSR |[√](https://github.com/LoSealL/Model/releases)|
        |DUF|[CVPR18](http://openaccess.thecvf.com/content_cvpr_2018/papers/Jo_Deep_Video_Super-Resolution_CVPR_2018_paper.pdf)|[Tensorflow](https://github.com/yhjo09/VSR-DUF)|T|-| VideoSR ||
        |CARN|[ECCV18](https://arxiv.org/abs/1803.08664)|[PyTorch](https://github.com/nmhkahn/CARN-pytorch)|Y|Y| Fast |[√](https://github.com/LoSealL/Model/releases/carn)|
        |RCAN|[ECCV18](https://arxiv.org/abs/1807.02758)|[PyTorch](https://github.com/yulunzhang/RCAN)|Y|Y| Deep, BI-BD-DN ||
        |MSRN|[ECCV18](http://openaccess.thecvf.com/content_ECCV_2018/papers/Juncheng_Li_Multi-scale_Residual_Network_ECCV_2018_paper.pdf)|[PyTorch](https://github.com/MIVRC/MSRN-PyTorch)|Y|Y| |[√](https://drive.google.com/open?id=1A0LoY3oB_VnArP3GzI1ILUNJbLAEjdtJ)|
        |SRFeat|[ECCV18](http://openaccess.thecvf.com/content_ECCV_2018/papers/Seong-Jin_Park_SRFeat_Single_Image_ECCV_2018_paper.pdf)|[Tensorflow](https://github.com/HyeongseokSon1/SRFeat)|Y|Y| GAN ||
        |NLRN|[NIPS18](https://papers.nips.cc/paper/7439-non-local-recurrent-network-for-image-restoration.pdf)|[Tensorflow](https://github.com/Ding-Liu/NLRN)|T|-| Non-local, Recurrent ||
        |SRCliqueNet|[NIPS18](https://arxiv.org/abs/1809.04508)|-|-|-| Wavelet ||
        |FFDNet|[TIP18](https://ieeexplore.ieee.org/document/8365806/)|[Matlab](https://github.com/cszn/FFDNet)|Y|Y| Conditional denoise||
        |CBDNet|[CVPR19](https://arxiv.org/abs/1807.04686)|[Matlab](https://github.com/GuoShi28/CBDNet)|T|-| Blind-denoise ||
        |SOFVSR|[ACCV18](http://arxiv.org/abs/1809.08573)|[PyTorch](https://github.com/LongguangWang/SOF-VSR)|-|Y| VideoSR |[√](https://github.com/LoSealL/Model/releases/download/sofvsr/SOFVSR_x4.zip)|
        |ESRGAN|[ECCVW18](http://arxiv.org/abs/1809.00219)|[PyTorch](https://github.com/xinntao/ESRGAN)|-|Y|1st place PIRM 2018|[√](https://github.com/LoSealL/Model/releases/download/esrgan/esrgan.zip)|
        |TecoGAN|[arXiv](http://arxiv.org/abs/1811.09393)|[Tensorflow](https://github.com/thunil/TecoGAN)|-|T| VideoSR GAN|[√](https://github.com/LoSealL/Model/releases/download/tecogan/tecogan.zip)|
        |RBPN|[CVPR19](https://arxiv.org/abs/1903.10128)|[PyTorch](https://github.com/alterzero/RBPN-PyTorch)|-|Y| VideoSR |[√](https://drive.google.com/open?id=1Ozp5j-DBWJSpXY5GvxiEPKdfCaAbOXqu)|
        |DPSR|[CVPR19](https://arxiv.org/abs/1903.12529)|[Pytorch](https://github.com/cszn/DPSR)|-|-|||
        |SRFBN|[CVPR19](https://arxiv.org/abs/1903.09814)|[Pytorch](https://github.com/Paper99/SRFBN_CVPR19)|-|-||||
        |SRNTT|[CVPR19](https://arxiv.org/abs/1903.00834)|[Tensorflow](https://github.com/ZZUTK/SRNTT)|-|-|Adobe||
        |SAN|[CVPR19](http://openaccess.thecvf.com/content_CVPR_2019/papers/Dai_Second-Order_Attention_Network_for_Single_Image_Super-Resolution_CVPR_2019_paper.pdf)|[empty](https://github.com/daitao/SAN)|-|-| AliDAMO SOTA ||
        |AdaFM|[CVPR19](https://arxiv.org/abs/1904.08118)|[Pytorch](https://github.com/hejingwenhejingwen/AdaFM)|-|-| SenseTime Oral ||
        
        \*The 1st repo is by paper author.
        
        \**__Y__: included; __-__: not included; __T__: under-testing. 
        
        You can download pre-trained weights through [`prepare_data`](./prepare_data.py), or visit the hyperlink at **√**.
        
        ## Link of datasets
        *(please contact me if any of links offend you or any one disabled)*
        
        |Name|Usage|#|Site|Comments|
        |:---|:----|:----|:---|:-----|
        |SET5|Test|5|[download](https://uofi.box.com/shared/static/kfahv87nfe8ax910l85dksyl2q212voc.zip)|[jbhuang0604](https://github.com/jbhuang0604/SelfExSR)|
        |SET14|Test|14|[download](https://uofi.box.com/shared/static/igsnfieh4lz68l926l8xbklwsnnk8we9.zip)|[jbhuang0604](https://github.com/jbhuang0604/SelfExSR)|
        |SunHay80|Test|80|[download](https://uofi.box.com/shared/static/rirohj4773jl7ef752r330rtqw23djt8.zip)|[jbhuang0604](https://github.com/jbhuang0604/SelfExSR)|
        |Urban100|Test|100|[download](https://uofi.box.com/shared/static/65upg43jjd0a4cwsiqgl6o6ixube6klm.zip)|[jbhuang0604](https://github.com/jbhuang0604/SelfExSR)|
        |VID4|Test|4|[download](https://people.csail.mit.edu/celiu/CVPR2011/videoSR.zip)|4 videos|
        |BSD100|Train|300|[download](https://uofi.box.com/shared/static/qgctsplb8txrksm9to9x01zfa4m61ngq.zip)|[jbhuang0604](https://github.com/jbhuang0604/SelfExSR)|
        |BSD300|Train/Val|300|[download](https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/segbench/BSDS300-images.tgz)|-|
        |BSD500|Train/Val|500|[download](http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/BSR/BSR_bsds500.tgz)|-|
        |91-Image|Train|91|[download](http://www.ifp.illinois.edu/~jyang29/codes/ScSR.rar)|Yang|
        |DIV2K|Train/Val|900|[website](https://data.vision.ee.ethz.ch/cvl/DIV2K/)|NTIRE17|
        |Waterloo|Train|4741|[website](https://ece.uwaterloo.ca/~k29ma/exploration/)|-|
        |MCL-V|Train|12|[website](http://mcl.usc.edu/mcl-v-database/)|12 videos|
        |GOPRO|Train/Val|33|[website](https://github.com/SeungjunNah/DeepDeblur_release)|33 videos, deblur|
        |CelebA|Train|202599|[website](http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html)|Human faces|
        |Sintel|Train/Val|35|[website](http://sintel.is.tue.mpg.de/downloads)|Optical flow|
        |FlyingChairs|Train|22872|[website](https://lmb.informatik.uni-freiburg.de/resources/datasets/FlyingChairs.en.html#flyingchairs)|Optical flow|
        |DND|Test|50|[website](https://noise.visinf.tu-darmstadt.de/)|Real noisy photos|
        |RENOIR|Train|120|[website](http://ani.stat.fsu.edu/~abarbu/Renoir.html)|Real noisy photos|
        |NC|Test|60|[website](http://demo.ipol.im/demo/125/)|Noisy photos|
        |SIDD(M)|Train/Val|200|[website](https://www.eecs.yorku.ca/~kamel/sidd/)|NTIRE 2019 Real Denoise|
        |RSR|Train/Val|80|[download]()|NTIRE 2019 Real SR|
        |Vimeo-90k|Train/Test|89800|[website](http://toflow.csail.mit.edu/)|90k HQ videos|
        
        Other open datasets:
        [Kaggle](https://www.kaggle.com/datasets)
        [ImageNet](http://www.image-net.org/)
        [COCO](http://cocodataset.org/)
        
        ## VSR package
        This package offers a training and data processing framework based on [TF](https://www.tensorflow.org).
        What I made is a simple, easy-to-use framework without lots of encapulations and abstractions.
        Moreover, VSR can handle raw NV12/YUV as well as a sequence of images as inputs.
        
        ### Install
        
        1. Prepare proper tensorflow and pytorch(optional). For example, GPU and CUDA10.0 (recommend to use `conda`):
        
           ```shell
           conda install tensorflow-gpu==1.15.0
           # optional
           # conda install pytorch
           ```
        
        2. Install VSR package
        
           ```bash
           # For someone see this doc online
           # git clone https://github.com/loseall/VideoSuperResolution && cd VideoSuperResolution
           pip install -e .
           ```
        ### Getting Started
        
        1. Download pre-trained weights and (optinal) training datasets. For instance, let\'s begin with VESPCN and vid4 test data:
           ```shell
           python prepare_data.py --filter vespcn vid4
           ```
           
        2. Customize backend
           cd ~/.vsr/
           touch config.yml
           ```yaml
           backend: tensorflow  # (tensorflow, pytorch)
           verbose: info        # (debug, info, warning, error)
           ```
        
        3. Evaluate
           ```shell
           cd Train
           python eval.py srcnn -t vid4 --pretrain=/path/srcnn.pth
           ```
        
        4. Train
           ```shell
           python prepare_data.py --filter mcl-v
           cd Train
           python train.py vespcn --dataset mcl-v --memory_limit 1GB --epochs 100
           ```
        
        OK, that's all you need. For more details, use `--help` to get more information.
        
        ----
        More documents can be found at [Docs](https://github.com/LoSealL/VideoSuperResolution/tree/master/Docs).
        
Keywords: super-resolution sr vsr cnn srcnn vespcn
Platform: UNKNOWN
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Requires-Python: >=3.6
Description-Content-Type: text/markdown
