ProGamerGov’s gists

ProGamerGov / original_colors.py

Last active July 2, 2018 19:05

A Python version of Neural-Style's color-independent style transfer function, using only PIL/Pillow.

	from PIL import Image

	import argparse
	parser = argparse.ArgumentParser()
	# Basic options
	parser.add_argument("-content", help="Original content target image", default='examples/inputs/tubingen.jpg')
	parser.add_argument("-generated", help="Stylized image", default='examples/outputs/tubingen_starry.png')
	parser.add_argument("-output_image", default='out.png')
	params = parser.parse_args()

ProGamerGov / neural_style_dir_rng_fix_fc.lua

Last active April 21, 2018 07:22

	-- This modified version of Neural-Style was created by github.com/ProGamerGov on 2017-09-29
	-- Use: "luarocks install paths" to install the required package.
	-- The -style_image parameter will now automatically determine if the input is a directory, or a string of one or more images.

	-- The FC layers as content layers feature comes from:
	-- github.com/htoyryla's gist.github.com/htoyryla/806ca4d978f0528114282cd00022ad71

	-- Code from: https://github.com/jcjohnson/neural-style/pull/408
	-- The NIN ImageNet model uses dropout layers with random values,
	-- so this version of neural_style.lua removes the unneeded dropout layers entirely.

ProGamerGov / neural_style_dir_rng_fix.lua

Last active April 19, 2018 20:27

	-- This modified version of Neural-Style was created by github.com/ProGamerGov on 2017-09-29
	-- Use: "luarocks install paths" to install the required package.
	-- The -style_image parameter will now automatically determine if the input is a directory, or a string of one or more images.

	-- Code from: https://github.com/jcjohnson/neural-style/pull/408
	-- The NIN ImageNet model uses dropout layers with random values,
	-- so this version of neural_style.lua removes the unneeded dropout layers entirely.

	require 'torch'
	require 'nn'

ProGamerGov / neural_bestchannels.lua

Last active February 14, 2018 22:44 — forked from htoyryla/neural_bestchannels.lua

	-- modified by ProGamerGov to select the best content channels, in addition to style channels.

	-- modified version of @jcjohnson's neural-style
	-- by @htoyryla 13 Feb 2018
	-- allows giving emphasis to nc best channel(s) in each style layer
	-- use -style_layers to select layer as usual, using a single layer is recommended
	-- -nc to set how many of the best channels are used per layer
	-- during target capture, tests the model using the style image
	-- and selects nc channels with strongest activations to be given emphasis during iterations
	-- not tested with multiple style images

ProGamerGov / convis.lua

Last active February 26, 2018 01:58

	-- Modified from github.com/htoyryla's code here: https://github.com/htoyryla/convis/issues/2#issuecomment-365009705

	require 'torch'
	require 'nn'
	require 'image'
	require 'loadcaffe'

	function preprocess(img)
	local mean_pixel = torch.DoubleTensor({103.939, 116.779, 123.68})
	local perm = torch.LongTensor{3, 2, 1}

ProGamerGov / neural_style_adam.lua

Last active November 15, 2017 21:47

	-- This version of neural_style.lua allows for the use of all of the Adam optimizer parameters.
	-- The learning rate parameter also now affect's L-BFGS' learning rate.

	-- The modifications were made based on the research done here:
	-- https://github.com/jcjohnson/neural-style/issues/428#issuecomment-342618091

	-- github.com/ProGamerGov 2017-11-07


	require 'torch'

ProGamerGov / neural_style_dir.lua

Last active September 30, 2017 21:16

This modified neural_style.lua allows for a directory to be used as an input for the -style_image parameter.

	-- This modified version of Neural-Style was created by github.com/ProGamerGov on 2017-09-29
	-- Use: "luarocks install paths" to install the required package.
	-- The -style_image parameter will now automatically determine if the input is a directory, or a string of one or more images.

	require 'torch'
	require 'nn'
	require 'image'
	require 'optim'
	require 'paths'

ProGamerGov / neural_style_rng_fix.lua

Last active September 24, 2017 21:50

The NIN ImageNet model uses dropout layers with random values, so this version of neural_style.lua removes the unneeded dropout layers entirely.

	-- Code from: https://github.com/jcjohnson/neural-style/pull/408
	-- The NIN ImageNet model uses dropout layers with random values,
	-- so this version of neural_style.lua removes the unneeded dropout layers entirely.

	require 'torch'
	require 'nn'
	require 'image'
	require 'optim'

	require 'loadcaffe'

ProGamerGov / alternate.sh

Last active September 22, 2020 21:04

Scripts that create rotated and reflected version of a specified image

	#! /bin/bash

	#Usage: ./alternate.sh <input.png>
	#Dependency: sudo apt-get install imagemagick
	#Permission fix troubleshooting: chmod u+x ./alternate.sh

	# 1. Defines the input image as a variable
	input=$1
	input_file=`basename $input`
	clean_name="${input_file%.*}"

ProGamerGov / neural_style_laplacian.lua

Last active June 11, 2017 23:38

Neural-Style with the laplacian feature from deep-photo-styletransfer. See the comments for setting up and using the new feature.

	-- Matting laplacian code from: github.com/martinbenson/deep-photo-styletransfer/
	-- Generate the laplacian with: https://gist.github.com/ProGamerGov/290f26afccc5e013d1a8425ef6a594f2
	-- More details can be found in this gist: https://gist.github.com/ProGamerGov/d3988b914c92220ae45aa5bb1e31d6a7
	-- The original Neural Style code can be found here: https://github.com/jcjohnson/neural-style

	require 'torch'
	require 'nn'
	require 'image'
	require 'optim'