Two people are attending a professional convention, likely focused on industrial or urban environments. The attendees are engaged in a display featuring urban scene generation with satellite images and diffused lighting. They are standing in front of an art display showcasing various urban elements, such as buildings, roads, and other features. Both individuals are wearing badges, indicating their participation in the event. The sign behind the attendees specifies that they should not use their cameras or phones to photograph the art, suggesting the presence of valuable or copyrighted material. Overall, the scene suggests a collaborative atmosphere and shared interest in the study and representation of urban environments.
Text transcribed from the image:
Highlight
Hzürich
kyo
uction
HEANG UNIVERSITY
OF AMSTERDAM
Sat2Scene: 3D Urban Scene Generation from Satellite Images with Diffusion
|浙江大学 Microsoft UNIVERSITY
Zuoyue Li Zhenqiang Li Zhaopeng Cui Marc Pollefeys Martin R. Oswald
enerate 3D urban scene o
bitrary 2D views with rob
WWWWZ
Al with Better Data Shape
VOXEL51
4 Experiment
metry and render
Baseline comparison Method/Metric | FVD
HoliCity dataset
GT geometry
Various metrics
Sat2 Vid
InfiniCity
MVDiffusion
37.06
KVDx 100+ FID
4.03 0.05
137.84
Ours
22.79
20.30
2.36±0.03
1.90 0.03
50.78
71.98
13.76±0.10
108.47 8.40±0.10
4.14±0.07
17.56 0.593
5.91±0.06 31.54
KID 100 PSNR SSIM1 LPIPS User study
Ablat
25.25 0.741 0.252
2.92%
⚫ w/o
w/o
0.259
15.62%
•
0.956
tion?
urally holds
set trajectory
models
ANS?
rmance
ring training
models
ethod
red
es
0.237 81.46%
w/o
w/o & w/ pc
Exemplary
scene used
for training
Variant/Metric
w/o pt-rsmp
w/o pt-aggr
w/o
dep-sup
Ours
GT
Sat2Vid
iffusion
Ours
GT
Sat2Vid
MVDiffusion
w/o pt-rsmp
w/o pt-aggr
w/o de
Model generalization OmniCity dataset, long-seq generation on
Sat.
Ground-view
Bird-view
MVDiff.
Ground-view
Mano
Sat2Scene: 3D Urban Scene Generation
from Satellite Images with Diffusion
OPF 204 SE
GT
Depths
Conclusio
ntributions
sparse diffe
grated with
realism a