Datasets:
gremlin97
/
RemoteSensingCorpus

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
820
results of this experiment. The red spot indicates the loca-
tion with the highest activations for the given query. For
each query, the left figure shows the total area of inference,
and the right figure shows a fine-grained image at the loca-
tion with the highest activation, obtained from Google. Wesee that our model is capable of making reasonable local-
ization for the given queries. For example: in (a) our model
activates over a soccer stadium. Similarly, for (c) and (d),
our model has high activations over an amusement park and
the “National Railway Museum” respectively. Figure (b)
shows that when we compose the concept of people with
animals, our model shows very high activation in farm-like
areas which is where these two concepts would most likely
co-occur. These results show that our model can reasonably
localize the most plausible point within a given area, where
one might observe a given query. This property can be ben-
eficial in solving visual search problems in the geospatial
domain.
5. Conclusion
We introduced a novel weakly supervised framework to
learn a rich embedding space between geolocation and fine-
grained captions. Our method does not require any text-
labeled data making it easy to train and scale. We demon-
strated 4 interesting applications of our model. First, we
showed that our model can be used for cross-view image
retrieval even when using uncurated ground-level images.
Secondly, we showed that our model can be used for gen-
erating fine-grained and dynamic captions for geolocations.
Third, we showed that our model can effectively localize
textual concepts within a given geospatial region. Finally,
we demonstrated how Sat2Cap embeddings can be used for
the newly defined task of large-scale zero-shot mapping.
8
References
[1] R. Mokady, A. Hertz, and A. H. Bermano, “Clip-
cap: Clip prefix for image captioning,” arXiv preprint
arXiv:2111.09734 , 2021.
[2] T. Salem, S. Workman, and N. Jacobs, “Learning a dynamic
map of visual appearance,” in Proceedings of the IEEE/CVF
Conference on Computer Vision and Pattern Recognition ,
pp. 12435–12444, 2020.
[3] P.-Y . Laffont, Z. Ren, X. Tao, C. Qian, and J. Hays, “Tran-
sient attributes for high-level understanding and editing of
outdoor scenes,” ACM Transactions on graphics (TOG) ,
vol. 33, no. 4, pp. 1–11, 2014.
[4] B. Zhou, A. Lapedriza, A. Khosla, A. Oliva, and A. Torralba,
“Places: A 10 million image database for scene recognition,”
IEEE transactions on pattern analysis and machine intelli-
gence , vol. 40, no. 6, pp. 1452–1464, 2017.
[5] A. Streltsov, J. M. Malof, B. Huang, and K. Bradbury, “Esti-
mating residential building energy consumption using over-
head imagery,” Applied Energy , vol. 280, p. 116018, 2020.
[6] A. J. Bency, S. Rallapalli, R. K. Ganti, M. Srivatsa, and
B. Manjunath, “Beyond spatial auto-regressive models: Pre-
dicting housing prices with satellite imagery,” in 2017
IEEE winter conference on applications of computer vision
(WACV) , pp. 320–329, IEEE, 2017.
[7] A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh,
S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, et al. ,
“Learning transferable visual models from natural language
supervision,” in International conference on machine learn-
ing, pp. 8748–8763, PMLR, 2021.
[8] J. Li, R. Selvaraju, A. Gotmare, S. Joty, C. Xiong, and
S. C. H. Hoi, “Align before fuse: Vision and language rep-
resentation learning with momentum distillation,” Advances
in neural information processing systems , vol. 34, pp. 9694–
9705, 2021.
[9] L. Yao, R. Huang, L. Hou, G. Lu, M. Niu, H. Xu,
X. Liang, Z. Li, X. Jiang, and C. Xu, “Filip: fine-grained
interactive language-image pre-training,” arXiv preprint
arXiv:2111.07783 , 2021.
[10] B. Thomee, D. A. Shamma, G. Friedland, B. Elizalde, K. Ni,
D. Poland, D. Borth, and L.-J. Li, “Yfcc100m: The new
data in multimedia research,” Communications of the ACM ,
vol. 59, no. 2, pp. 64–73, 2016.
[11] L. Ilic, M. Sawada, and A. Zarzelli, “Deep mapping gentrifi-
cation in a large canadian city using deep learning and google
street view,” PloS one , vol. 14, no. 3, p. e0212814, 2019.
[12] T. Behrens, K. Schmidt, R. A. MacMillan, and R. A. Vis-
carra Rossel, “Multi-scale digital soil mapping with deep
learning,” Scientific reports , vol. 8, no. 1, p. 15244, 2018.
[13] Z. Zong, J. Feng, K. Liu, H. Shi, and Y . Li, “Deepdpm: Dy-
namic population mapping via deep neural network,” in Pro-
ceedings of the AAAI Conference on Artificial Intelligence ,
vol. 33, pp. 1294–1301, 2019.[14] M. Onishi and T. Ise, “Explainable identification and map-
ping of trees using uav rgb image and deep learning,” Scien-
tific reports , vol. 11, no. 1, p. 903, 2021.
[15] C. Greenwell, S. Workman, and N. Jacobs, “What goes
where: Predicting object distributions from above,” in
IGARSS 2018-2018 IEEE International Geoscience and Re-
mote Sensing Symposium , pp. 4375–4378, IEEE, 2018.
[16] V . Alhassan, C. Henry, S. Ramanna, and C. Storie, “A deep
learning framework for land-use/land-cover mapping and
analysis using multispectral satellite imagery,” Neural Com-
puting and Applications , vol. 32, pp. 8529–8544, 2020.
[17] K. Karra, C. Kontgis, Z. Statman-Weil, J. C. Mazzariello,
M. Mathis, and S. P. Brumby, “Global land use / land cover
with sentinel 2 and deep learning,” in 2021 IEEE Interna-
tional Geoscience and Remote Sensing Symposium IGARSS ,
pp. 4704–4707, 2021.
[18] B. Feizizadeh, D. Omarzadeh, M. Kazemi Garajeh, T. Lakes,