demo.py

import os
import sys
import fnmatch

#Caffe binary location from deeplab-v1 installation
caffe_binary = '/vision/vision_users/suyog/deep_video_segmentation/code_release/deeplab-public/distribute/bin/caffe.bin' 

#Extension of images that need to be processed
ext = 'JPEG'

base_dir = os.getcwd()
image_dir = os.path.join(base_dir,'images')

image_list = fnmatch.filter(os.listdir(image_dir),'*.'+ext)
image_list.sort()

input_list_file  = base_dir + '/image_list.txt'
output_list_file = base_dir + '/output_list.txt'

input_list  = open(input_list_file,'w')
output_list = open(output_list_file,'w')

for img in image_list:
	input_list.write('/'+img+'\n')
	prefix = img.split('.')[0]
	output_list.write(prefix+'\n')

input_list.close()
output_list.close()

template_file = open(base_dir + '/test_template.prototxt').readlines()

test_file_path = base_dir + '/test.prototxt'
test_file = open(test_file_path,'w')

tokens = {}
tokens['${IMAGE_DIR}'] = 'root_folder: \"' + image_dir + '\"'
tokens['${OUTPUT_DIR}'] = 'prefix: \"' + image_dir + '/\"'

tokens['${IMAGE_LIST}']        = 'source: \"' + input_list_file + '\"'
tokens['${IMAGE_OUTPUT_LIST}'] = 'source: \"' + output_list_file + '\"'

for line in template_file:
	line = line.rstrip()

	for key in tokens:
		if line.find(key)!=-1:
			line = '\t'+tokens[key]
			break

	test_file.write(line+'\n')

test_file.close()

weight_file_path = base_dir + '/pixel_objectness.caffemodel'
cmd = caffe_binary + ' test --model=' + test_file_path + ' --weights=' + weight_file_path + ' --gpu=0 --iterations='+str(len(image_list))
print cmd
os.system(cmd)