antordragon Goto Github PK
Type: User
Type: User
2019CCF-BDCI大赛 最佳创新探索奖获得者 基于OCR身份证要素提取赛题冠军 天晨破晓团队 赛题源码
This project is to implement “Attention-Adaptive and Deformable Convolutional Modules for Dynamic Scene Deblurring(with ERCNN)” . To run this project you need to setup the environment, download the dataset, and then you can train and test the network models. ## Prerequiste The project is tested on Ubuntu 16.04, GPU Titan XP. Note that one GPU is required to run the code. Otherwise, you have to modify code a little bit for using CPU. If using CPU for training, it may too slow. So I recommend you using GPU strong enough and about 12G RAM. ## Dependencies Python 3.5 or 3.6 are recommended. ``` tqdm==4.19.9 numpy==1.17.3 torch==1.0.0 Pillow==6.1.0 torchvision==0.2.2 ``` ## Environment I recommend using ```virtualenv``` for making an environment. If you using ```virtualenv```, ## Dataset I use GOPRO dataset for training and testing. __Download links__: [GOPRO_Large](https://drive.google.com/file/d/1H0PIXvJH4c40pk7ou6nAwoxuR4Qh_Sa2/view?usp=sharing) | Statistics | Training | Test | Total | | ----------- | -------- | ---- | ----- | | sequences | 22 | 11 | 33 | | image pairs | 2103 | 1111 | 3214 | After downloading dataset successfully, you need to put images in right folders. By default, you should have images on dataset/train and dataset/valid folders. ## Demo ## Training Run the following command ``` python demo_train.py ('data_dir' is needed before running ) ``` For training other models, you should uncommend lines in scripts/train.sh file. I used ADAM optimizer with a mini-batch size 16 for training. The learning rate is 1e-4. Total training takes 600 epochs to converge. To prevent our network from overfitting, several data augmentation techniques are involved. In terms of geometric transformations, patches are randomly rotated by 90, 180, and 270 degrees. To take image degradations into account, saturation in HSV colorspace is multiplied by a random number within [0.8, 1.2]. ![validation_curves](figs/validation_curve.png) ## Testing Run the following command ``` python demo_test.py ('data_dir' is needed before running ) ``` ## pretrained models if you need the pretrained models,please contact us by [email protected] ## Acknowledge Our code is based on Deep Multi-scale Convolutional Neural Network for Dynamic Scene Deblurring [MSCNN](http://openaccess.thecvf.com/content_cvpr_2017/papers/Nah_Deep_Multi-Scale_Convolutional_CVPR_2017_paper.pdf), which is a nice work for dynamic scene deblurring .
Contrastive Learning for Compact Single Image Dehazing, CVPR2021
Domain Adaptation for Image Dehazing, CVPR2020
Enhanced Pix2pix Dehazing Network, accepted by CVPR 2019
ICRA 2019 "FastDepth: Fast Monocular Depth Estimation on Embedded Systems"
FFA-Net: Feature Fusion Attention Network for Single Image Dehazing
Inference pipeline for the CVPR paper entitled "Real-Time Monocular Depth Estimation using Synthetic Data with Domain Adaptation via Image Style Transfer" (http://www.atapour.co.uk/papers/CVPR2018.pdf).
Official repository for "Multi-Stage Progressive Image Restoration" (CVPR 2021). SOTA results for image deblurring, deraining, and denoising.
The source code of CVPR 2020 paper "Multi-Scale Boosted Dehazing Network with Dense Feature Fusion"
NBNet: Noise Basis Learning for Image Denoising with Subspace Projection
PyTorch implementation of Deformable ConvNets v2 (Modulated Deformable Convolution)
official implementation of "Revisiting Single Image Depth Estimation: Toward Higher Resolution Maps with Accurate Object Boundaries"
Repository for Scale-recurrent Network for Deep Image Deblurring
U-Net: Convolutional Networks for Biomedical Image Segmentation
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.