GitPlanet

Cheap and reliable Node.js hosting starts at $3/month, and $1/month static HTML hosting

Created with love in Canada, visit hostnodejs.com today

Feel like to post an Ad? Learn Details
All Projects → LeeDoYup → Deblurgan Tf

LeeDoYup / Deblurgan Tf

Licence: mit
Unofficial tensorflow (tf) implementation of DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks

Programming Languages

python
139335 projects - #7 most used programming language

Labels

deep-learningtensorflowcnngan

Projects that are alternatives of or similar to Deblurgan Tf

Keraspp
코딩셰프의 3분 딥러닝, 케라스맛
Stars: ✭ 178 (+196.67%)
Mutual labels:  cnn, gan
DLSS
Deep Learning Super Sampling with Deep Convolutional Generative Adversarial Networks.
Stars: ✭ 88 (+46.67%)
Mutual labels:  cnn, gan
Cyclegan Music Style Transfer
Symbolic Music Genre Transfer with CycleGAN
Stars: ✭ 201 (+235%)
Mutual labels:  cnn, gan
Visual Feature Attribution Using Wasserstein Gans Pytorch
Implementation of Visual Feature Attribution using Wasserstein GANs (VAGANs, https://arxiv.org/abs/1711.08998) in PyTorch
Stars: ✭ 88 (+46.67%)
Mutual labels:  cnn, gan
Basicocr
BasicOCR是一个致力于解决自然场景文字识别算法研究的项目。该项目由长城数字大数据应用技术研究院佟派AI团队发起和维护。
Stars: ✭ 336 (+460%)
Mutual labels:  cnn, gan
Tensorflow Tutorials
텐서플로우를 기초부터 응용까지 단계별로 연습할 수 있는 소스 코드를 제공합니다
Stars: ✭ 2,096 (+3393.33%)
Mutual labels:  cnn, gan
Srgan
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
Stars: ✭ 2,641 (+4301.67%)
Mutual labels:  cnn, gan
Anime Face Gan Keras
A DCGAN to generate anime faces using custom mined dataset
Stars: ✭ 161 (+168.33%)
Mutual labels:  cnn, gan
Dcpdn
Densely Connected Pyramid Dehazing Network (CVPR'2018)
Stars: ✭ 321 (+435%)
Mutual labels:  cnn, gan
Pytorch Srgan
A modern PyTorch implementation of SRGAN
Stars: ✭ 289 (+381.67%)
Mutual labels:  cnn, gan
Segan
A PyTorch implementation of SEGAN based on INTERSPEECH 2017 paper "SEGAN: Speech Enhancement Generative Adversarial Network"
Stars: ✭ 82 (+36.67%)
Mutual labels:  cnn, gan
Tensorflow Tutorial
Tensorflow tutorial from basic to hard, 莫烦Python 中文AI教学
Stars: ✭ 4,122 (+6770%)
Mutual labels:  cnn, gan
Iseebetter
iSeeBetter: Spatio-Temporal Video Super Resolution using Recurrent-Generative Back-Projection Networks | Python3 | PyTorch | GANs | CNNs | ResNets | RNNs | Published in Springer Journal of Computational Visual Media, September 2020, Tsinghua University Press
Stars: ✭ 202 (+236.67%)
Mutual labels:  cnn, gan
Mydeeplearning
A deep learning library to provide algs in pure Numpy or Tensorflow.
Stars: ✭ 281 (+368.33%)
Mutual labels:  cnn, gan
Time Series Prediction
A collection of time series prediction methods: rnn, seq2seq, cnn, wavenet, transformer, unet, n-beats, gan, kalman-filter
Stars: ✭ 351 (+485%)
Mutual labels:  cnn, gan
Deeplearning
深度学习入门教程, 优秀文章, Deep Learning Tutorial
Stars: ✭ 6,783 (+11205%)
Mutual labels:  cnn, gan
Facenet Face Recognition
This is the research product of the thesis manifold Learning of Latent Space Vectors in GAN for Image Synthesis. This has an application to the research, name a facial recognition system. The application was developed by consulting the FaceNet model.
Stars: ✭ 54 (-10%)
Mutual labels:  gan
Lstm Context Embeddings
Augmenting word embeddings with their surrounding context using bidirectional RNN
Stars: ✭ 57 (-5%)
Mutual labels:  cnn
Lung Diseases Classifier
Diseases Detection from NIH Chest X-ray data
Stars: ✭ 52 (-13.33%)
Mutual labels:  cnn
Image captioning
generate captions for images using a CNN-RNN model that is trained on the Microsoft Common Objects in COntext (MS COCO) dataset
Stars: ✭ 51 (-15%)
Mutual labels:  cnn
View All Similar Projects ➔

DeblurGAN-tf

DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks paper
This repository is tensorflow (tf) implementation of DeblurGAN.

DeblurGAN removes blur filter in a image and make the image sharp, as follow:

We use tensorflow 1.7.0 and opencv-python 3.4.0.12 and only support CPU version or single GPU.

Notice

  • I had reproduce the results, but someone mentioned they didn't have satisfactory results. (If you also reproduce, please share experience for everyone.)
  • DeblurGAN-v2 is launched (in ICCV'19). code

First, Download GOPRO Dataset

DeblurGAN model is trained using GOPRO dataset.
To train the model, download the dataset from website download

Unzip the dataset wherever you want, and remember the (data_path).

Model Training

You can train DeblurGAN model from the sratch, using GOPRO dataset.
(Also, you can train if you have your own blur/sharp pair images) .

python3 train.py --data_path_train (training data path) --data_path_test (test data path)

There are many argument options and you can check them in train.py .
When you use custom dataset, please set some arguments:

  • --resize_or_crop (default: resize) : resize or crop the input image to (img_h, img_w, img_c)
  • --img_h (int, default: 256)
  • --img_w (int, default: 256)
  • --img_c (int, default: 256)

When the model training ends, you can identify the results of Debluring GOPRO dataset.
The results are saved in ./test_result/

You can also change the checkpoint and tensorboard directory using args.

  • --checkpoint_dir (str, default: ./checkpoints)
  • --model_name (str, default: DeblurGAN.model)
  • --summary_dir (str, default: ./summaries)

Tensorboard

During training, tensorboard is available.

tensorboard --logdir=(summary path, default: './summaries') --port=(port number, default:8888)

Loss Values

Model Graph

If you want to monitor intermediate results in training process, you can see the deblur image in tensorboard with additional option.

  • --tf_image_monitor (bool, default: False)

Deblur your own Images

After training the model, you can deblur your own images using the trained model.
First, your own images have to be in same directory and same filename extension (png, jpg, or etc...).

python test.py --is_test --data_path_test (your data directory path) --img_type (image filename extension)

Two argumnets are important:

  • --data_path_test (str, default: None) : the your own data directory path have to be entered.
  • --result_dir(str, default: ./result_dir) : the dirtectory path where the sharp images are saved.

Etc

According to the original paper, model training takes about 6 days using NVIDIA TITAN X.
To increase training speed, i am writing tf.data modules to feed the data without bottleneck.
Pull request and issues are always welcome.

Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].