import requests from requests.adapters import HTTPAdapter from requests.adapters import Retry import json import os import concurrent.futures import argparse from datetime import datetime import writer from model import PictureType import sys session = requests.Session() retries_strategies = Retry( total=5, backoff_factor=1, status_forcelist=[429,502, 503, 504], ) session.mount('https://', HTTPAdapter(max_retries=retries_strategies)) def parse_args(argv =None): parser = argparse.ArgumentParser() parser.add_argument('access_token', type=str, help='Your mapillary access token') parser.add_argument('--sequence_ids', type=str, nargs='*', help='The mapillary sequence id(s) to download') parser.add_argument('--image_ids', type=int, nargs='*', help='The mapillary image id(s) to get their sequence id(s)') parser.add_argument('--destination', type=str, default='data', help='Path destination for the images') parser.add_argument('--image_limit', type=int, default=None, help='How many images you want to download') parser.add_argument('--overwrite', default=False, action='store_true', help='overwrite existing images') parser.add_argument("-v", "--version", action="version", version="release 1.2") args = parser.parse_args(argv) if args.sequence_ids is None and args.image_ids is None: parser.error("Please enter at least one sequence id or image id") return args def download(url, filepath, metadata=None): #print(asizeof.asizeof(image)/1024, "MB") with open(str(filepath), "wb") as f: r = session.get(url, stream=True, timeout=6) image = write_exif(r.content, metadata) f.write(image) print("{} downloaded".format(filepath)) def get_single_image_data(image_id, mly_header): req_url = 'https://graph.mapillary.com/{}?fields=thumb_original_url,altitude,camera_type,captured_at,compass_angle,geometry,exif_orientation,sequence'.format(image_id) r = session.get(req_url, headers=mly_header) data = r.json() #print(data) return data def get_image_data_from_sequences(sequences_id, mly_header): for i,sequence_id in enumerate(sequences_id): url = 'https://graph.mapillary.com/image_ids?sequence_id={}'.format(sequence_id) r = requests.get(url, headers=header) data = r.json() image_ids = data['data'] total_image = len(image_ids) print("{} images in sequence {} of {} - id : {}".format(total_image, i+1, len(sequences_id), sequence_id)) print('getting images data') for x in range(0, total_image): image_id = image_ids[x]['id'] image_data = get_single_image_data(image_id, mly_header) image_data['sequence_id'] = sequence_id yield image_data def get_image_data_from_sequences__future(sequences_id, mly_header): for i,sequence_id in enumerate(sequences_id): url = 'https://graph.mapillary.com/image_ids?sequence_id={}'.format(sequence_id) r = requests.get(url, headers=header) data = r.json() if data.get('data') == []: print("Empty or wrong sequence {} of {} - id : {}".format(i+1, len(sequences_id), sequence_id)) continue image_ids = data['data'] total_image = len(image_ids) print("{} images in sequence {} of {} - id : {}".format(total_image, i+1, len(sequences_id), sequence_id)) print('getting images data') with concurrent.futures.ThreadPoolExecutor(max_workers=10) as executor: future_to_url = {} for x in range(0, total_image): image_id = image_ids[x]['id'] future_to_url[executor.submit(get_single_image_data, image_id, mly_header)] = image_id for future in concurrent.futures.as_completed(future_to_url): url = future_to_url[future] image_data = future.result() image_data['sequence_id'] = sequence_id #print(image_data) yield image_data def write_exif(picture, img_metadata): ''' Write exif metadata ''' #{'thumb_original_url': 'https://scontent-cdg4-2.xx.fbcdn.net/m1/v/t6/An9Zy2SrH9vXJIF01QkBODyUbg7XSKfwL48UwHyvihSwvECGjVbG0vSw9uhxe2-Dq-k2eUcigb83buO6zo-7eVbykfp5aQIe1kgd-MJr66nU_H-o_mwBLZXgVbj5I_5WX-C9c6FxJruHkV962F228O0?ccb=10-5&oh=00_AfDOKD869DxL-4ZNCbVo8Rn29vsc0JyjMAU2ctx4aAFVMQ&oe=65256C25&_nc_sid=201bca', # 'captured_at': 1603459736644, 'geometry': {'type': 'Point', 'coordinates': [2.5174596904057, 48.777089857534]}, 'id': '485924785946693'} with writer.Writer(picture) as image: image.add_datetimeoriginal(img_metadata) image.add_lat_lon(img_metadata) image.add_altitude(img_metadata) image.add_direction(img_metadata) image.add_img_projection(img_metadata) image.apply() updated_image = image.get_Bytes() return updated_image if __name__ == '__main__': args = parse_args() sequence_ids= args.sequence_ids if args.sequence_ids is not None else [] images_ids = args.image_ids access_token = args.access_token images_data = [] header = {'Authorization' : 'OAuth {}'.format(access_token)} if images_ids: for image_id in images_ids: image_data = get_single_image_data(image_id, header) if 'error' in image_data: print("data : ", image_data) print("something wrong happened ! Please check your image id and/or your connection") sys.exit() else: sequence_ids.append(image_data.get('sequence')) #for i,image_data in enumerate(get_image_data_from_sequences(sequence_ids, header)): for i,image_data in enumerate(get_image_data_from_sequences__future(sequence_ids, header)): if args.image_limit is not None and i >= args.image_limit: break if 'error' in image_data: print("data : ", image_data) print("something wrong happened ! Please check your token and/or your connection") sys.exit() images_data.append(image_data) #sys.exit() print('downloading.. this process will take a while. please wait') with concurrent.futures.ThreadPoolExecutor(max_workers=4) as executor: for i,image_data in enumerate(images_data): # create a folder for each unique sequence ID to group images by sequence path_destination = os.path.join(args.destination, image_data['sequence_id']) if not os.path.exists(path_destination): os.makedirs(path_destination) date_time_image_filename = datetime.utcfromtimestamp(int(image_data['captured_at'])/1000).strftime('%Y-%m-%d_%HH%Mmn%Ss%f')[:-3] + '.jpg' path = os.path.join(path_destination, date_time_image_filename) img_metadata = writer.PictureMetadata( capture_time = datetime.utcfromtimestamp(int(image_data['captured_at'])/1000), longitude = image_data['geometry']['coordinates'][0], latitude = image_data['geometry']['coordinates'][1], picture_type = PictureType("equirectangular") if image_data['camera_type'] == 'spherical' or image_data['camera_type'] == 'equirectangular' else PictureType("flat"), direction = image_data['compass_angle'], altitude = image_data['altitude'], ) image_exists = os.path.exists(path) if not args.overwrite and image_exists: print("{} already exists. Skipping ".format(path)) continue executor.submit(download, url=image_data['thumb_original_url'], filepath=path, metadata=img_metadata) #download(url=image_data['thumb_original_url'], filepath=path, metadata=img_metadata)