End-to-end network for blur detection and classification (presented at PG 2018)
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
input
model
output
README.md
config.py
main.py
model.py
utils.py

README.md

deep_blur_detection_and_classification

Tensorflow implementation of "Defocus and Motion Blur Detection with Deep Contextual Features"

For image examples:

input2 output2

This repository contains a test code and sythetic dataset, which consists of scenes including motion and defocus blurs together in each scene.


Prerequisites

  • Ubuntu 16.04
  • Tensorflow 1.6.0 or higher
  • Tensorlayer 1.8.2 or higher
  • OpenCV2

Test Details

  • download model weights from google drive and save the model into 'model' folder.
  • specify a path of input folder in 'main.py' at line #39
  • run 'main.py'
python main.py

Synthetic Dataset

  • download synthetic train set(337MB) and synthetic test set(11.5MB) from google drive
  • Note that sharp pixels, motion-blurred pixels, and defocus-blurred pixels in GT blur maps are labeled as 0, 100, and 200, respectively, in the [0,255] range.