jureviciusr / gamd-net

Gyroscope-Aided Motion Deblurring with Deep Networks

Our results on Blur-IMU

Synthetic blurred image

Comparison of different methods

Composite blurred image

Prerequisites

• Python
• Scipy
• Pandas
• OS
• math
• opencv

Installation

Clone this project to your machine.

git clone https://github.com/lsmlovefm/GAMD-Net.git
cd Image_blur

Process

Download blurred image metadata from dataset inside Imageblur/data. Then place the image that needs to be blurred into imageblur/image. It should be noted that the default image pixels are 740*580. You can use this method to combine clear images in Realblur and GoPro into blurry images.

Imageblur
├─ Imageblur
│ ├─ blur    % blurred image
│ │ ├─ xxxx.jpg
│ │ ├─ ......
│ │
│ ├─ image % raw image
│ │ ├─ xxxx.jpg
│ │ ├─ ......
│
│ ├─ data % download from [this](https://drive.google.com/file/d/10inB3MHqycfK1awgBy13lNNyj1L8amvY/view?usp=drive_link)

python imageblur/imageblur.py

GAMD network

Prerequisites

• Python
• Pytorch (1.4)
• scikit-image
• opencv-python
• Tensorboard

Installation

Clone this project to your machine.

git clone https://github.com/lsmlovefm/GAMD-Net.git
cd GAMD

Process

Download blurred image metadata from IMU-blur inside GAMD/.

GAMD
├─ trainingset
│ ├─ train
│ │ ├─ blur    % 6680 image pairs
│ │ │ ├─ xxxx.png
│ │ │ ├─ ......
│ │ │
│ │ ├─ sharp
│ │ │ ├─ xxxx.png
│ │ │ ├─ ......
│ │ 
│ │ ├─ cp % control point heatmap
│ │ │ ├─ xxxx.png
│ │ │ ├─ ......
│ │
│ │ ├─ ep % endpoint heatmap
│ │ │ ├─ xxxx.png
│ │ │ ├─ ......
│ │
│ ├─ test    % 1670 image pairs
│ │ ├─ ...... (same as train)

Train

Train GAMD, run the command below:

python main.py --mode "train" --data_dir "training_set"

Model weights will be saved in results/model_name/weights folder.

Test

Test GAMD , run the command below:

python main.py --mode "test" --data dir "training_set" --test_model "best.pkl"

Model link: https://drive.google.com/file/d/1Z-Af3qCtzJZAoOyfDItCYDPssVs_Wx8S/view?usp=drive_link