Description
The development of this data set aims at testing and improving deep learning algorithms for fisheye image. The original images of the fisheye data set are chosen from the image database ImageNet.
Characteristic
Fisheye lens could cover over 180 degree fields of view with heavy edge distortion. When the optical axis angle of a fisheye lens changes, a great distortion change happens on a specific object in two images. To simulate distortions on different view angels and view points, different transformation offset are chosen. A example for a eagle image is shown as above.
Synthetic sequences
n02676566
acoustic_guitar
n02747177
ashcan
n02783161
ballpoint
n02870880
bookcase
n02992529
cellular_telephone
n03017168
chime
n03041632
cleaver
n03063689
coffeepot
n03179701
desk
n03180011
desktop_computer
n03207743
dishrag
n03207941
dishwasher
n03297495
espresso_maker
n0348331
hand_blower
n03529860
home_theater
n03637318
lampshade
n03991062
pot
n04004767
printer
n04009552
projector
n04070727
refrigerator
n04099969
rocking_chair
n04204347
shopping_cart
n04209239
shower_curtain
n04326547
stone_wal
n04330267
stove
n04367480
swab
n04398044
teapot
n04404412
television
n04442312
toaster
n04447861
toilet_seat
n04517823
vacuum
n04548280
wall_clock
n04550184
wardrobe
n04553703
washbasin
n04557648
water_bottle
n04560804
water_jug
n04590129
window_shade
n04596742
wok
n07248320
book_jacket
n07892512
red_wine
Licenses & Credits
The data set is publically, freely, and exclusively available for research purposes.
A compilation of the image files and blender models used for the synthetic sequences along with their respective licensing information is provided in this document.