三次元形状処理とディープラーニングの初歩についてまとめたスライドです。
View Slide
•機械学習ニューラルネット深層学習
•• • θ• = ; = 22 + 1 + 0෩ = arg min =0−1(, ; )
•••0= 0 + 01= 1 + 12= 2 + 2 = 00+ 11+ 22+ = 00+ 11+ 22+(00+ 11+ 22+ )
•••0120= ReLU(0 + 0)1= ReLU(1 + 1)2= ReLU(2 + 2)
= 32 = 128 = 512••
•• = 512 = 1 = = 4
•
•••••••••••
(c) Velodyne(c) Andrew Tallon(c) Microsoft(c) Sony
• , , = 0
′ = − ′ = + ′
•☺☺☺☺ , , = 0
☺☺☺☺ , , = 0
•••••
••••
•⚫⚫⚫
•••
••
•••••••
•••(c) Velodyne
•••(0, 0, 0)(1, 1, 1)(, , )(−1, −1, −1)(+1, +1, +1)(0, 0, 0)(1, 1, 1)(−1, −1, −1)(+1, +1, +1)(, , )
⚫
•• ∗ = ℱ−1(ℱ ⋅ ℱ )′ = ℱ−1(Θ ∘ ℱ )
• =0 1 1 01 0 1 01 1 0 10 0 1 0 = − −12−12= :
•• • = T
• ()′ = ℱ−1(Θ ∘ ℱ )′ = ℱ = ()ℱ−1 = ()
•••+1+2+1 = ReLU ⋅
••••••
• , , =
••••••••
•••••••••
tatsy
ad5jp
miyakemito
oracle4engineer
uzulla
orgachem
tarao
drumath2237
takufujii
tomohisa
sakaik
megabitsenmzq
makamaka
frogandcode
maggiecrowley
reverentgeek
paulrobertlloyd
tanoku
addyosmani
paigeccino
iamctodd
michaelherold
lara
pedronauck
marcduiker