Seanforfun / GMAN_Net_Haze_Removal

Single Image Dehazing with a Generic Model-Agnostic Convolutional Neural Network
https://ieeexplore.ieee.org/document/8686264
38 stars 21 forks source link

Hazy Image generation machanism #2

Closed Seanforfun closed 6 years ago

Seanforfun commented 6 years ago
  1. How to get a depth of a scene? And can we get a matting graph with real distance but not fractions.
  2. How Resides database generate their outdoor and indoor images. @MintcakeDotCom
Seanforfun commented 6 years ago

Indoor images are using depth map get from Kinect, depth are not only the values from 0 to 1, but estimated real value.

Seanforfun commented 6 years ago

How RESIDES generate hazy images

Outdoor

Mentioned in Resides Paper

What They really use

Indoor(已验证,也合成了一样的有雾图)

How to create synthetic hazed image

1.使用已经存在的深度图片集(NYU,用于室内)或算法(用于室外)得到深度的估计值,单位是米。 2.随机取出beta和alpha值,均在0到1之间。 3.通过t = e^(-beta * depth)得到深度图,值域在0到1之间。 4.通过I(有雾图) = J(无雾图)* t + (1 - t) * alpha得到合成图。