My first video deep-dream

export_video_inceptionism.py

from batcountry import BatCountry
import numpy as np
from PIL import Image
from glob import glob
import os
import random

CAFFE_ROOT = '../caffe'
INPUT_PATH = 'input.jpg'

def gradient_ascent_step(net, step_size=1.5, end="inception_4c/output",
		jitter=32, clip=True, objective_fn=None, **objective_params):
		# if the objective function is None, initialize it as
		# the standard L2 objective
		if objective_fn is None:
			objective_fn = BatCountry.L2_objective

		# input image is stored in Net's 'data' blob
		src = net.blobs["data"]
		dst = net.blobs[end]

		# apply jitter shift
		ox, oy = np.random.randint(-jitter, jitter + 1, 2)

		print ox, oy
	
		src.data[0] = np.roll(np.roll(src.data[0], ox, -1), oy, -2)

		net.forward(end=end)
		objective_fn(dst, **objective_params)
		net.backward(start=end)
		g = src.diff[0]

		# apply normalized ascent step to the input image
		src.data[:] += step_size / np.abs(g).mean() * g # origin

 		#unshift image
		src.data[0] = np.roll(np.roll(src.data[0], -ox, -1), -oy, -2)

		# unshift image
		if clip:
			bias = net.transformer.mean["data"]
			src.data[:] = np.clip(src.data, -bias, 255 - bias)

def dream(path):
	print '%s/models/bvlc_googlenet' % (CAFFE_ROOT)
	bc = BatCountry('%s/models/bvlc_googlenet' % (CAFFE_ROOT))

	image = bc.dream(np.float32(Image.open(path)), end='inception_3a/output', step_fn=gradient_ascent_step, step_size=2, octave_n=12)
	bc.cleanup()

	filename = 'result/perlin_c/out_%s' % os.path.basename(path)

	result = Image.fromarray(np.uint8(image))
	result.save(filename)

for path in glob('./perlin_c/*.png'):
	# print "BK: processing... %s" % path
	np.random.seed(1234)
	dream(path)