/HairNet_orient2D

This repo contains the code for generating the orientation images in the paper HairNet: Single-View Hair Reconstruction using Convolutional Neural Networks.

Primary LanguageC++MIT LicenseMIT

#
#  Created by Yi 16 June 2020
#

#This repo contains the code for generating the orientation images in the paper HairNet: Single-View Hair Reconstruction using Convolutional Neural Networks.


#####################################STEP ONE##################################################################
#Compile the code in Hair_Orient2D.

#Please check the CMakeLists for dependencies.

#OPENGL, GLEW, GLFW3 and OpenMesh are only used for rendering a white body silhouette. If you don't need it, you can simply remove them and compile with the optional main_nobodyrender.cpp.

#####################################STEP Two##################################################################
#Compute 2D orienation maps from portrait images in given folder.

#Put the test imgs in "test_imgs/img/", the test images should be square, e.g. 800*800
#Manually draw the body image and the hair segments. (T_T Sorry, I can't opensource this part so please either manually draw it or look for some other opensource codes.)
#Put the body images in "test_imgs/body_img/", with the same name as in img folder, XXX.png. body is white, background is black
#Put the segment images in "test_imgs/seg/", with the same name as in img folder, XXX.png. hair is white, background is black
#Use command:
Hair_Orient2D/Orient2D 1 test_imgs/
#argv: has_body_img, hair_folder

#if you somehow can retreat the transformation matrix of the head in the image as described in our paper, you can also choose to automatically render the body as follows:
#Put the transformation matrices(4x4) in "test_imgs/exMat/"
#Use command:
Hair_Orient2D/Orient2D 0 test_imgs/


#Note: all the directory arguments need to have "/" at the end of the string.
#I put some sample input and output in "test_img/".