Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
ABEJA Innovation Meetup NIPS PointNet++
Search
望月紅葉さんと幸せな家庭を築きたい
January 01, 2018
Programming
1
480
ABEJA Innovation Meetup NIPS PointNet++
望月紅葉さんと幸せな家庭を築きたい
January 01, 2018
Tweet
Share
More Decks by 望月紅葉さんと幸せな家庭を築きたい
See All by 望月紅葉さんと幸せな家庭を築きたい
shadow-detection-with-conditional-generative-adversarial-networks
momijifullmoon
0
140
unsupervised-learning-of-depth-and-ego-motion-from-monocular-video-using-3d-geometric-constraints
momijifullmoon
0
380
NIPS2017reading_3Dreconstruction
momijifullmoon
0
1.5k
Other Decks in Programming
See All in Programming
Amazon Neptuneで始めてみるグラフDB-OpenSearchによるグラフの全文検索-
satoshi256kbyte
4
330
Realtime API 入門
riofujimon
0
110
Macとオーディオ再生 2024/11/02
yusukeito
0
210
[PyCon Korea 2024 Keynote] 커뮤니티와 파이썬, 그리고 우리
beomi
0
110
qmuntal/stateless のススメ
sgash708
0
120
CPython 인터프리터 구조 파헤치기 - PyCon Korea 24
kennethanceyer
0
250
『ドメイン駆動設計をはじめよう』のモデリングアプローチ
masuda220
PRO
8
440
Boost Performance and Developer Productivity with Jakarta EE 11
ivargrimstad
0
880
Progressive Web Apps für Desktop und Mobile mit Angular (Hands-on)
christianliebel
PRO
0
110
Vue SFCのtemplateでTypeScriptの型を活用しよう
tsukkee
3
1.5k
GCCのプラグインを作る / I Made a GCC Plugin
shouth
1
150
Universal Linksの実装方法と陥りがちな罠
kaitokudou
1
220
Featured
See All Featured
Fashionably flexible responsive web design (full day workshop)
malarkey
404
65k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
280
13k
Imperfection Machines: The Place of Print at Facebook
scottboms
264
13k
Designing Dashboards & Data Visualisations in Web Apps
destraynor
228
52k
Large-scale JavaScript Application Architecture
addyosmani
510
110k
Speed Design
sergeychernyshev
24
570
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
31
2.7k
Unsuck your backbone
ammeep
668
57k
Happy Clients
brianwarren
97
6.7k
10 Git Anti Patterns You Should be Aware of
lemiorhan
654
59k
Reflections from 52 weeks, 52 projects
jeffersonlam
346
20k
Templates, Plugins, & Blocks: Oh My! Creating the theme that thinks of everything
marktimemedia
26
2.1k
Transcript
PointNet++: Deep Hierarchical Feature Learning on Point Sets
in a Metric Space NIPSಡΈձˏABEJA 1
PointNet++ͷ֓ཁ ▸ ஶऀ: Charles R. Qi, Li Yi, Hao Su,
Leonidas J. Guibas ɹɹ ˏελϯϑΥʔυ ▸ ֓ཁ ▸ ܈Λͦͷ··ೖྗ͠ɺͦͷΫϥεྨɺ SegmentationΛߦ͏PointNetͷվྑใࠂ ▸ PointNetͷऑͰ͋ͬͨ܈ີґଘΛࠀɺ ͓Αͼ֊తͳֶशΛͰ͖ΔΑ͏ʹ ʮSampling Layerʯͱ ʮGrouped LayerʯΛఏҊ 2
എܠ ▸ ̏࣍ݩͷधཁ 3 ࣗಈӡస AR ઃܭ
ͷྲྀΕ ▸ എܠ ▸ PointNetʹ͍ͭͯ ▸ ख๏ ▸ ࣮ݧ ▸
·ͱΊ 4
എܠ ▸ ̏࣍ݩͷσʔλ 5 ɹɹ܈ɹɹ ɹɹϝογϡɹɹ Voxel Өɹ RGB-D
എܠ ▸ طଘͷख๏ ▸ ܈Λผͷදݱʹม͍ͯͨ͠ 6 Unstructured, Unordered ͳ܈Λͦͷ··ೖྗ Ͱ͏·͍͘͘Α͏ͳख๏
==> PointNetΛఏҊ@CVPR2017
PointNetͷ͓͞Β͍ ▸ ղ͘λεΫ 7 Classification Segmentation Scene Parsing ೖྗ
PointNetͷ͓͞Β͍ ▸ ઃܭ 8
PointNetͷ͓͞Β͍ ▸ ՝ 9 PointNet֤ʹ͓͍ͯɺlocalͷใ͕ফ͑Δ ֊తಛֶशͰ͖ͳ͍ ෳ֊ͷநԽͰ͖ͳ͍ GlobalͷಛֶशͷΈ ͋Δ͘͠શͯͷ
PointNetͷ͓͞Β͍ ▸ localͷใ͕ফ͑Δͱ 10 globalͷใɺઈର࠲ඪʹґଘͯ͠͠·͏ͷͰɺ segmentationͰະͷͷʹରԠͰ͖ͳ͍
PointNet++Ͱ ▸ ֊తֶश ▸ localͳใΛ͢ 11 ▸ ܈ີʹϩόετʹ
ΞʔΩςΫνϟ 12
֊తͳֶश 13
֊తͳֶश ▸ Sampling layer ▸ Farthest Point Sampling (FPS) 14
https://www.groundai.com/project/parametric-manifold-learning-via-sparse-multidimensional-scaling/
▸ Grouping layer ▸ radius based ball query ֊తͳֶश 15
PointNet layer Convolution layer Input Δԋࢉ ԋࢉͰݟΔ ൣғ Radius ball query ɹ܈ɹ PointNetʢॱ൪ීวʣ ߦྻʢݻఆͷϐΫηϧʣ ΈࠐΈʢॱ൪ґଘʣ ɹີͳߦྻɹ
֊తͳֶश ▸ PointNet layer 16 N1ݸͷʹର͠ C1ݸͷಛ࡞ ॏΈshare
֊తͳֶश ▸ PointNet layer 17 x1,y1,z1,ΫΤϦ1,ಛ1 x2,y2,z2,ΫΤϦ2,ಛ2 x3,y3,z3,ΫΤϦ3,ಛ3 xN1,yN1,zN1,ΫΤϦN1ಛN1 MLP
MLP MLP MLP x1,y1,z1,ಛ1 x2,y2,z2,ಛ2 x3,y3,z3,ಛ3 xN1,yN1,zN1,ಛN1 ॏΈShare
ີґଘࠀख๏ ▸ ̏࣍ݩͷଌఆͰ܈ີ͕Ұൠతͳ՝ 18 ==> ܈ີʹϩόετʹ͍ͨ͠
ີґଘࠀख๏ ▸ SamplingͱGroupingΛෳ༻ҙ 19 MRGͷํ͕࣍ͰपลͱͷಛΛर͑Δ
Classification ࣮ݧ 20
▸ ModelNet40ʹରͯ͠ Classification ࣮ݧ 21 PointNetʹൺɺPointNet++ྨਫ਼্ CNNϕʔεͷख๏ʹউར
ີґଘ࣮ݧ 22 ಛʹ܈͕গͳ͍ͱɺMRG͕༗ޮ
Segmentation ࣮ݧ 23 ૠɿɹIDW (ٯڑՃॏ) Unitpointnet: ֤ͰMLP
Segmentation ࣮ݧ ▸ ݁Ռ 24 MSGΛೖΕΔ͜ͱͰɺෆۉҰͳ܈Ͱ͏·͍͘͘
Segmentation ࣮ݧ ▸ ݁Ռ 25 PointNetΑΓՈ۩ͷsegmentation্͕ख͍͘͘
ඇϢʔΫϦου ڑۭؒͰͷ࣮ݧ 26 WKS , HKS, multi-scale Gaussian curvature
Feature Visualization ▸ ࠷ॳͷͷॏΈΛՄࢹԽ 27 ฏ໘ɺίʔφʔͱ͔Λֶश
·ͱΊ ▸ PointNetΛ֦ுͨ͠ख๏PointNet++Λൃද ▸ CVPR2017=>NIPS2017ʹ̍ຊ௨͍ͯ͠Δɻɻɻ ▸ Sampling layerɺGrouped layerΛऔΓೖΕ֊తͳֶश ▸
MRGɺMSGΛఏҊ͠ɺ܈ີʹґଘ͠ͳֶ͍श ▸ ̏࣍ݩ܈ͷσʔληοτʹରͯ͠ɺSoTAୡ ▸ ݱʹߦͬͨײ ▸ ஶऀͱ͢͜ͱͰࡉ͔ͳใΛर͑Δ 28