Using Data Sources¶

Streetscapes can work with images from multiple sources. Currently included are: Mapillary, Kartaview, and Amsterdam Open Panorama.

It is also possible to use Global Streetscapes as a data source. However, as it is a fundamentally different type of dataset, a separate tutorial can be found here

NOTE: Make sure you have the appropriate API keys stored in your .env file.

• Mapillary
• Kartaview
• Amsterdam Open Panorama

First let's load the libraries and a workspace:

In [1]:

import warnings
warnings.filterwarnings("ignore")

import geopandas as gpd
import contextily
from PIL import Image

import numpy as np
from matplotlib import pyplot as plt

from streetscapes.sources import Mapillary, KartaView, AmsterdamPanorama
from streetscapes.streetview import SVWorkspace

import warnings warnings.filterwarnings("ignore") import geopandas as gpd import contextily from PIL import Image import numpy as np from matplotlib import pyplot as plt from streetscapes.sources import Mapillary, KartaView, AmsterdamPanorama from streetscapes.streetview import SVWorkspace

In [2]:

ws = SVWorkspace("./Amsterdam", create=True)

ws = SVWorkspace("./Amsterdam", create=True)

Mapillary¶

Mapillary is an open source platform for sharing crowdsourced geotagged images, with different capturing modes including walking, cycling and driving. This is one of the most popular alternatives to Google Streetview.

First, load the data source into your workspace environment:

In [3]:

mp = Mapillary(ws.env)
mp

mp = Mapillary(ws.env) mp

Out[3]:

Mapillary(root_dir='~/.cache/huggingface/assets/streetscapes/images/mapillary')

To test that the connection to our data sources work, we can get the image URL for an image:

In [4]:

mp_img_url = mp.get_image_url("4911230068985425")
mp_img_url

mp_img_url = mp.get_image_url("4911230068985425") mp_img_url

Out[4]:

'https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/An--rrF4WlM05JH5SMiWxbdSXVWL0mFEymquJWW0EmSjaBemmKBcF8DRGqIe8akNayRiH0VTYunIfeHZlt2zAGA0sYkEkeWIUtnfp27cgiCen37fra1_v0gwTJFgNL5wTiUGkS7F56f_yLBZF1HqZg?stp=s2048x1536&edm=ALXxkZ8EAAAA&_nc_gid=tRbdW7qYYq0X5shRFZKzqQ&_nc_oc=Adkq7sMroKF2dZYGjwd90Vy8lKypHmU9SUEHP5gqwLGIVCHE_qxvwlQePNnu-8C4UOIzdVfTHVpZYuM7NdLV1_YK&ccb=10-5&oh=00_AfIS3HNjILFUkNkGg0xxkyr0hbenPw-1Q-YFarbRfKXC-Q&oe=68668C7C&_nc_sid=201bca'

Then we can fetch all of the image ids within a certain bounding box (bbox). You can define the bbox you want here.

In [5]:

bbox = [4.879990,52.368023,4.892499,52.374656]
df = mp.fetch_image_ids(bbox=bbox)

bbox = [4.879990,52.368023,4.892499,52.374656] df = mp.fetch_image_ids(bbox=bbox)

We can then take these ids and turn them to a geopandas dataframe and visualise them:

In [6]:

pdf = df.to_pandas()
gdf = gpd.GeoDataFrame(pdf, geometry=gpd.points_from_xy(pdf.lon, pdf.lat))
gdf.set_crs("EPSG:4326", allow_override=True, inplace=True)
gdf.head(5)

pdf = df.to_pandas() gdf = gpd.GeoDataFrame(pdf, geometry=gpd.points_from_xy(pdf.lon, pdf.lat)) gdf.set_crs("EPSG:4326", allow_override=True, inplace=True) gdf.head(5)

Out[6]:

	id	altitude	atomic_scale	camera_type	captured_at	compass_angle	computed_altitude	computed_compass_angle	computed_geometry	computed_rotation	...	make	model	thumb_256_url	thumb_1024_url	thumb_2048_url	thumb_original_url	sequence	width	lon	lat
0	1917152201975344	50.157000	0.966447	fisheye	1683058662000	233.678855	2.200463	234.444802	{'coordinates': [4.8890121741514, 52.368506077...	[0.80341789785712, 1.5825194278947, -1.6909058...	...	GoPro	HERO7 Black	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	hEqPyRUmOBlvVJkbpzMg3d	4000	4.889012	52.368506
1	2123857951153524	59.137000	0.958617	fisheye	1683058745000	302.127134	1.332355	300.733279	{'coordinates': [4.8881621068794, 52.368739030...	[1.3840344032714, 0.87375049055148, -0.7304818...	...	GoPro	HERO7 Black	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	hEqPyRUmOBlvVJkbpzMg3d	4000	4.888162	52.368739
2	1905249926497728	84.565000	1.025931	fisheye	1683222429000	194.520986	1.624200	169.813657	{'coordinates': [4.890530623151, 52.3729970104...	[0.30890820372818, -2.1646534079862, 2.1842481...	...	GoPro	HERO7 Black	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	NhvZLRndBupJwiQCjV8D75	4000	4.890531	52.372997
3	3894119180663616	46.585438	1.001886	equirectangular	1471422392024	0.000000	2.186527	0.095146	{'coordinates': [4.8821141861186, 52.370025099...	[1.5755633925048, 0.00038934569121215, 0.00299...	...	Trimble	Trimble TMX	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	cjXHBMe3gjtnO1JxY0itMg	8000	4.882114	52.370025
4	4026155874166112	50.994701	0.999110	equirectangular	1471238655950	0.000000	2.074441	1.326067	{'coordinates': [4.8914778199576, 52.370309523...	[1.5715531011154, -0.0045901262039534, 0.03175...	...	Trimble	Trimble TMX	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	https://scontent-ams4-1.xx.fbcdn.net/m1/v/t6/A...	a9oQejn_GKWe73KqIZrcSQ	8000	4.891478	52.370310

5 rows × 24 columns

In [7]:

ax = gdf.plot(column='uuid', color='red', markersize=3, figsize=(6, 6), alpha=0.5)
contextily.add_basemap(ax, crs=gdf.crs, source=contextily.providers.nlmaps.standaard)

ax = gdf.plot(column='uuid', color='red', markersize=3, figsize=(6, 6), alpha=0.5) contextily.add_basemap(ax, crs=gdf.crs, source=contextily.providers.nlmaps.standaard)

Finally, we can download an image:

In [8]:

image = gdf.iloc[0]
filename = mp.download_image(image.id, image.thumb_1024_url, overwrite=True)
Image.open(filename)

image = gdf.iloc[0] filename = mp.download_image(image.id, image.thumb_1024_url, overwrite=True) Image.open(filename)

Out[8]:

For multiple images, the process is similar:

In [9]:

df_sample = gdf[:5] # First 5 images
image_ids = df_sample.id # Column in the dataframe with id
urls = df_sample.thumb_1024_url # Specify the type of url

paths = mp.download_images(image_ids, urls)
paths

df_sample = gdf[:5] # First 5 images image_ids = df_sample.id # Column in the dataframe with id urls = df_sample.thumb_1024_url # Specify the type of url paths = mp.download_images(image_ids, urls) paths

Downloading images | mapillary | Done: 100%|██████████| 5/5 [00:02<00:00,  2.00it/s]                

Out[9]:

[PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/mapillary/1917152201975344.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/mapillary/2123857951153524.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/mapillary/1905249926497728.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/mapillary/3894119180663616.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/mapillary/4026155874166112.jpeg')]

KartaView¶

KartaView, formerly OpenStreetView and OpenStreetCam is a project to crowdsource street-level photographs to improve OpenStreetMap. It works quite similarly to Mapillary but unlike Mapillary, it also provides some open source code of its web and mobile applications. The coverage of KartaView is a little bit more restricted with fewer images depending on the region.

Let's load the data source into the workspace environment:

In [10]:

kv = KartaView(ws.env)
kv

kv = KartaView(ws.env) kv

Out[10]:

KartaView(root_dir='~/.cache/huggingface/assets/streetscapes/images/kartaview')

To test that the connection to our data sources work, we can get the image URL for an image:

In [11]:

kv_img_url = kv.get_image_url("1208524")
kv_img_url

kv_img_url = kv.get_image_url("1208524") kv_img_url

Unlike Mapillary, KartaView downloads images within a certain radius of a co-ordinate. Here we have chosen a 500 m radius:

In [12]:

df = kv.fetch_image_ids(52.3723032, 4.8877233, 500)

df = kv.fetch_image_ids(52.3723032, 4.8877233, 500)

We can then take these ids and turn them to a geopandas dataframe and visualise them:

In [13]:

pdf = df.to_pandas()
gdf = gpd.GeoDataFrame(pdf, geometry=gpd.points_from_xy(pdf.lng, pdf.lat))
gdf.set_crs("EPSG:4326", allow_override=True, inplace=True)
gdf.head(5)

pdf = df.to_pandas() gdf = gpd.GeoDataFrame(pdf, geometry=gpd.points_from_xy(pdf.lng, pdf.lat)) gdf.set_crs("EPSG:4326", allow_override=True, inplace=True) gdf.head(5)

Out[13]:

	id	sequence_id	sequence_index	orgCode	lat	lng	field_of_view	name	lth_name	th_name	...	match_lat	match_lng	way_id	shot_date	heading	headers	gps_accuracy	projection	username	geometry
0	1605307145	6072841	5218	CMNT	52.369301	4.892199	None	storage13/files/photo/2022/9/10/proc/6072841_5...	storage13/files/photo/2022/9/10/lth/6072841_58...	storage13/files/photo/2022/9/10/th/6072841_58a...	...	None	None	None	2022-08-20 09:29:56.000	15.18	None	None	PLANE	madflex	POINT (4.8922 52.3693)
1	1605307401	6072841	5233	CMNT	52.370079	4.892559	None	storage13/files/photo/2022/9/10/proc/6072841_6...	storage13/files/photo/2022/9/10/lth/6072841_6b...	storage13/files/photo/2022/9/10/th/6072841_6b4...	...	52.370079040527344	4.892553806304932	7371712	2022-08-20 09:30:12.000	190.80	None	None	PLANE	madflex	POINT (4.89256 52.37008)
2	1605307657	6072841	5247	CMNT	52.370702	4.892724	None	storage13/files/photo/2022/9/10/proc/6072841_6...	storage13/files/photo/2022/9/10/lth/6072841_64...	storage13/files/photo/2022/9/10/th/6072841_64d...	...	52.370700836181640	4.892742156982422	7371712	2022-08-20 09:30:26.000	13.72	None	None	PLANE	madflex	POINT (4.89272 52.3707)
3	1605307913	6072841	5265	CMNT	52.371511	4.892822	None	storage13/files/photo/2022/9/10/proc/6072841_b...	storage13/files/photo/2022/9/10/lth/6072841_bc...	storage13/files/photo/2022/9/10/th/6072841_bc3...	...	52.371509552001950	4.892897129058838	7371712	2022-08-20 09:30:44.000	5.43	None	None	PLANE	madflex	POINT (4.89282 52.37151)
4	1605308169	6072841	5280	CMNT	52.372230	4.892816	None	storage13/files/photo/2022/9/10/proc/6072841_d...	storage13/files/photo/2022/9/10/lth/6072841_d3...	storage13/files/photo/2022/9/10/th/6072841_d32...	...	52.372226715087890	4.892929077148438	7371712	2022-08-20 09:30:59.000	3.38	None	None	PLANE	madflex	POINT (4.89282 52.37223)

5 rows × 23 columns

In [14]:

ax = gdf.plot(column='uuid', color='red', markersize=3, figsize=(6, 6), alpha=0.5)
contextily.add_basemap(ax, crs=gdf.crs, source=contextily.providers.nlmaps.standaard)

ax = gdf.plot(column='uuid', color='red', markersize=3, figsize=(6, 6), alpha=0.5) contextily.add_basemap(ax, crs=gdf.crs, source=contextily.providers.nlmaps.standaard)

Finally, we can download an image:

In [15]:

image = gdf.iloc[0]
url = kv.get_image_url(image)
filename = kv.download_image(image.id, url, overwrite=True)
Image.open(filename)

image = gdf.iloc[0] url = kv.get_image_url(image) filename = kv.download_image(image.id, url, overwrite=True) Image.open(filename)

Out[15]:

For multiple images, the process is similar:

In [16]:

df_sample = gdf[:5] # First 5 images
image_ids = df_sample.id # Column in the dataframe with id
urls = [kv.get_image_url(image) for image in image_ids]

paths = kv.download_images(image_ids, urls)
paths

df_sample = gdf[:5] # First 5 images image_ids = df_sample.id # Column in the dataframe with id urls = [kv.get_image_url(image) for image in image_ids] paths = kv.download_images(image_ids, urls) paths

Downloading images | kartaview | Done: 100%|██████████| 5/5 [00:00<00:00, 879.97it/s]                 

Out[16]:

[PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/kartaview/1605307145.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/kartaview/1605307401.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/kartaview/1605307657.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/kartaview/1605307913.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/kartaview/1605308169.jpeg')]

Amsterdam Open Panorama¶

Amsterdam Open Panorama is project by the city of Amsterdam to provide up to date 360 degrees street view panoramas. These are updated several times a year.

Let's load the data source into our workspace environment:

In [17]:

aop = AmsterdamPanorama(ws.env)

aop = AmsterdamPanorama(ws.env)

Like KartaView, Amsterdam Open Panorama downloads images for a radius around a coordinate. Here we use 100 m as there are a lot of images.

In [24]:

df = aop.fetch_image_ids(52.3723032, 4.8877233, 100)

df = aop.fetch_image_ids(52.3723032, 4.8877233, 100)

We can then take these ids and turn them to a geopandas dataframe and visualise them:

In [25]:

pdf = df.to_pandas()
gdf = gpd.GeoDataFrame(pdf, geometry=gpd.points_from_xy(pdf.lon, pdf.lat))
gdf.set_crs("EPSG:4326", allow_override=True, inplace=True)
gdf.head(5)

pdf = df.to_pandas() gdf = gpd.GeoDataFrame(pdf, geometry=gpd.points_from_xy(pdf.lon, pdf.lat)) gdf.set_crs("EPSG:4326", allow_override=True, inplace=True) gdf.head(5)

Out[25]:

	pano_id	timestamp	lon	lat	height	heading	roll	pitch	thumbnail_url	cubic_img_baseurl	cubic_img_pattern	equirectangular_full	geometry
0	TMX7316010203-002865_pano_0017_000042	2022-12-15T12:25:13.423810Z	4.887726	52.372259	46.602516	95.884121	-0.422495	-0.740015	https://api.data.amsterdam.nl/panorama/thumbna...	https://t1.data.amsterdam.nl/panorama/2022/12/...	https://t1.data.amsterdam.nl/panorama/2022/12/...	https://t1.data.amsterdam.nl/panorama/2022/12/...	POINT (4.88773 52.37226)
1	TMX7316010203-002865_pano_0017_000043	2022-12-15T12:25:14.593580Z	4.887660	52.372261	46.607472	95.102385	0.269215	-0.316865	https://api.data.amsterdam.nl/panorama/thumbna...	https://t1.data.amsterdam.nl/panorama/2022/12/...	https://t1.data.amsterdam.nl/panorama/2022/12/...	https://t1.data.amsterdam.nl/panorama/2022/12/...	POINT (4.88766 52.37226)
2	TMX7316010203-002865_pano_0017_000041	2022-12-15T12:25:12.368630Z	4.887793	52.372256	46.606823	95.415160	-1.313683	-0.901538	https://api.data.amsterdam.nl/panorama/thumbna...	https://t1.data.amsterdam.nl/panorama/2022/12/...	https://t1.data.amsterdam.nl/panorama/2022/12/...	https://t1.data.amsterdam.nl/panorama/2022/12/...	POINT (4.88779 52.37226)
3	TMX7316010203-002537_pano_0001_000805	2022-04-22T08:32:02.226790Z	4.887712	52.372234	46.982223	274.629359	0.458672	-0.064371	https://api.data.amsterdam.nl/panorama/thumbna...	https://t1.data.amsterdam.nl/panorama/2022/04/...	https://t1.data.amsterdam.nl/panorama/2022/04/...	https://t1.data.amsterdam.nl/panorama/2022/04/...	POINT (4.88771 52.37223)
4	TMX7316010203-002445_pano_0002_000021	2022-01-27T08:40:48.378890Z	4.887696	52.372231	46.446957	95.441268	1.141474	-0.285900	https://api.data.amsterdam.nl/panorama/thumbna...	https://t1.data.amsterdam.nl/panorama/2022/01/...	https://t1.data.amsterdam.nl/panorama/2022/01/...	https://t1.data.amsterdam.nl/panorama/2022/01/...	POINT (4.8877 52.37223)

In [26]:

ax = gdf.plot(column='uuid', color='red', markersize=3, figsize=(6, 6), alpha=0.5)
contextily.add_basemap(ax, crs=gdf.crs, source=contextily.providers.nlmaps.standaard)

ax = gdf.plot(column='uuid', color='red', markersize=3, figsize=(6, 6), alpha=0.5) contextily.add_basemap(ax, crs=gdf.crs, source=contextily.providers.nlmaps.standaard)

Finally, we can download an image:

In [27]:

image = gdf.iloc[0]
filename = aop.download_image(image.pano_id, image.thumbnail_url, overwrite=True)
Image.open(filename)

image = gdf.iloc[0] filename = aop.download_image(image.pano_id, image.thumbnail_url, overwrite=True) Image.open(filename)

Out[27]:

The process is similar for multiple images:

In [28]:

df_sample = gdf[:5] # First 5 images
image_ids = df_sample.pano_id # Column in the dataframe with id
urls = df_sample.thumbnail_url

paths = aop.download_images(image_ids, urls)
paths

df_sample = gdf[:5] # First 5 images image_ids = df_sample.pano_id # Column in the dataframe with id urls = df_sample.thumbnail_url paths = aop.download_images(image_ids, urls) paths

Downloading images | amsterdampanorama | Done: 100%|██████████| 5/5 [00:00<00:00, 1328.74it/s]                                 

Out[28]:

[PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/amsterdampanorama/TMX7316010203-002865_pano_0017_000042.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/amsterdampanorama/TMX7316010203-002865_pano_0017_000043.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/amsterdampanorama/TMX7316010203-002865_pano_0017_000041.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/amsterdampanorama/TMX7316010203-002537_pano_0001_000805.jpeg'),
 PosixPath('/Users/clairedonnelly/.cache/huggingface/assets/streetscapes/images/amsterdampanorama/TMX7316010203-002445_pano_0002_000021.jpeg')]