readdata.py

# -*- coding: utf-8 -*-
import cv2
import numpy as np
import pdb

def load_split_nvgesture(file_with_split = './nvgesture_train_correct.lst',list_split = list()):
	params_dictionary = dict()
	with open(file_with_split,'rb') as f:
		dict_name  = file_with_split[file_with_split.rfind('/')+1 :]
		dict_name  = dict_name[:dict_name.find('_')]

		for line in f:
			params = line.split(' ')
			params_dictionary = dict()

			params_dictionary['dataset'] = dict_name

			path = params[0].split(':')[1]
			for param in params[1:]:
				parsed = param.split(':')
				key = parsed[0]
				if key == 'label':
					# make label start from 0
					label = int(parsed[1]) - 1 
					params_dictionary['label'] = label
				elif key in ('depth','color','duo_left'):
					#othrwise only sensors format: <sensor name>:<folder>:<start frame>:<end frame>
					sensor_name = key
					#first store path
					params_dictionary[key] = path + '/' + parsed[1]
					#store start frame
					params_dictionary[key+'_start'] = int(parsed[2])

					params_dictionary[key+'_end'] = int(parsed[3])
        
			params_dictionary['duo_right'] = params_dictionary['duo_left'].replace('duo_left', 'duo_right')
			params_dictionary['duo_right_start'] = params_dictionary['duo_left_start']
			params_dictionary['duo_right_end'] = params_dictionary['duo_left_end']          

			params_dictionary['duo_disparity'] = params_dictionary['duo_left'].replace('duo_left', 'duo_disparity')
			params_dictionary['duo_disparity_start'] = params_dictionary['duo_left_start']
			params_dictionary['duo_disparity_end'] = params_dictionary['duo_left_end']                  

			list_split.append(params_dictionary)
 
	return list_split

def load_data_from_file(example_config, sensor,image_width, image_height):

	path = example_config[sensor] + ".avi"
	start_frame = example_config[sensor+'_start']
	end_frame = example_config[sensor+'_end']
	label = example_config['label']

	frames_to_load = range(start_frame, end_frame)

	chnum = 3 if sensor == "color" else 1

	video_container = np.zeros((image_height, image_width, chnum, 80), dtype = np.uint8)

	cap = cv2.VideoCapture(path)

	ret = 1
	frNum = 0
	cap.set(cv2.CAP_PROP_POS_FRAMES, start_frame);
	for indx, frameIndx in enumerate(frames_to_load):    
		ret, frame = cap.read()
		if ret:
			frame = cv2.resize(frame,(image_width, image_height))
			if sensor != "color":
				frame = frame[...,0]
				frame = frame[...,np.newaxis]
			video_container[..., indx] = frame
		else:
			print "Could not load frame"
            
	cap.release()

	return video_container, label

if __name__ == "__main__":
	sensors = ["color", "depth", "duo_left", "duo_right", "duo_disparity"]
	file_lists = dict()
	file_lists["test"] = "./nvgesture_test_correct.lst"
	file_lists["train"] = "./nvgesture_train_correct.lst"
	train_list = list()
	test_list = list()

	load_split_nvgesture(file_with_split = file_lists["train"],list_split = train_list)
	load_split_nvgesture(file_with_split = file_lists["test"],list_split = test_list)

	data, label = load_data_from_file(example_config = train_list[0], sensor = sensors[0], image_width = 320, image_height = 240)
	pdb.set_trace()