Datasets:
gremlin97
/
RemoteSensingCorpus

Dataset card Data Studio Files Community
1
text
stringlengths
0
820
data labels are not available. [9] rotates an image and then
trains a model to predict the rotation angle. [48] trains a
network to perform colorization of a grayscale image. [26]
represents an image as a grid, permuting the grid and then
predicting the permutation index. In this study, we use geo-location classification as a pre-text task, in which a deep
network is trained to predict a coarse geo-location of where
in the world the image might come from.
Contrastive Learning Recent self-supervised contrastive
learning approaches such as MoCo [13], MoCo-v2 [3], Sim-
CLR [2], PIRL [22], and FixMatch [32] have demonstrated
superior performance and have emerged as the fore-runner
on various downstream tasks. The intuition behind these
methods are to learn representations by pulling positive
image pairs from the same instance closer in latent space
while pushing negative pairs from difference instances fur-
ther away. These methods, on the other hand, differ in the
type of contrastive loss, generation of positive and negative
pairs, and sampling method.
Although growing rapidly in self-supervised learning
area, contrastive learning methods have not been explored
on large-scale remote sensing dataset. In this work, we pro-
vide a principled and effective approach for improving rep-
resentation learning using MoCo-v2 [13] for remote sensing
data as well geo-located conventional datasets.
Unsupervised Learning in Remote Sensing Images Un-
like in traditional computer vision areas, unsupervised
learning on remote sensing domain has not been studied
comprehensively. Most of the studies utilize small-scale
datasets specific to a small geographical region [4, 17, 30,
15, 19], a few classes [35, 25] or a highly-specific modal-
ity, i.e. hyperspectral images [23, 47]. Most of these
studies focus on the UCM-21 dataset [45] consisting of
less than 1,000 images from 21 classes. A more recent
study [36] proposes large-scale weakly supervised learn-
ing using a multi-modal dataset consisting of satellite im-
ages and paired geo-located wikipedia articles. While be-
ing effective, this method requires each satellite image to
be paired to its corresponding article, limiting the number
of images that can be used.
Geography-aware Computer Vision Geo-location data
has been studied extensively in prior works. Most of these
2
"gsd": "img_width": "img_height": "country_code": "cloud_cover": “timestamp": 2.10264849663 2421 2165 IND 6 2015-11-02 T05:44:14Z2.06074237823 2410 2156 IND 0 2016-03-09 T05:25:30Z1.9968634 2498 2235 IND 1 2017-02-02 T05:47:02Z2.2158575 2253 2015 IND 0 2017-02-27 T05:24:30Z1.24525177479 4016 3592 IND 0 2015-04-09 T05:36:04Z1.4581833 3400 3041 IND 2 2016-12-28 T05:57:06Z1.2518295 4003 3581 IND 0 2017-04-12 T05:51:49ZFigure 2: Images over time concept in the fMoW dataset. The metadata associated with each image is shown underneath.
We can see changes in contrast, brightness, cloud cover etc. in the images. These changes render spatially aligned images
over time useful for constructing additional positives.
-45.829337,
170.730800,
Te Matai,
New Zealand,
Otago10.595525,
76.041355,
Guruvayoor ,
India,
Kerala,
Thrissur29.008740,
-81.963415,
Lake W eir,
Florida,Marion
United States28.318038,
-81.540688,
Celebration,
United States,
Florida,Osceola21.881018,
-102.275360,
Aguascalientes,
México,
Aguascalientes,
Aguascalientes59.61 1778,
-151.449966,
Homer ,
United States,
Alaska,
Kenai
Peninsula Borough37.345989,
-77.645680,
Beach,
United States,
Virginia,
Chesterfield37.303518,
-121.897773,
San Jose,
United States,
California,
Santa Clara22.494622,
114.028447,
⼤浪,
⾹港,
新界,
元朗區27.876676,
-82.798231,
Walsingham,
United States,
Florida,
Pinellas
-18.646245,
24.614868,
Botswana,
Chobe
Figure 3: Some examples from GeoImageNet dataset. Below each image, we list their latitudes, longitudes, city, country
name. In our study, we use the latitude and longitude information for unsupervised learning. We recommend readers to