Upgrade to Pro — share decks privately, control downloads, hide ads and more …

輪講資料 / SelFlow: Self-Supervised Learning of Opt...

輪講資料 / SelFlow: Self-Supervised Learning of Optical Flow

2019/10/18 社内 AIシステム部CVチームの輪講にて発表させていただいた資料です

Takumi Karasawa

October 18, 2019
Tweet

More Decks by Takumi Karasawa

Other Decks in Research

Transcript

  1. o 2 C A F C nw hP -A (

    G S R ir T - C O 0 9 nw ugU 2 C A F C 0-1 sp ugv • - - ltU - C )1 mpU 0 9 • T A 9 C 9 C V aLbecdf lt A ) M1 C GN
  2. 2 - 1- M N C • 3 - 232

    3 - 4 - O 3 M N • 3 - 3 - 4 - I 3 M N S M N 3 - C - 4- -
  3. + 34 4 4 B + 4 1 4 -.)

    4 0 1 4 FaMN W ] l MP M • [ km • A1 3 AF C 1 A1 3 AF c eh M ehfi d . - .
  4. ( ) - 1 - 1 - - 5 C

    5 ( 5 [ WN Nac - C S] 5 5 P 1 - C C
  5. ( ) - 16- 6 1 d - - S

    a W 6 S 6 6 ( ] 6 S[ P Pce - S 6 6 C 6 6 S 6 N , 2
  6. - 1 3 9 . 3 229 P a .

    9 C O TL 229 1 U F 229 3 3 9O M 2 Ud M . 3 O[] 2 E 2 1 N C1 3 1 P c
  7. -- -3 - -= -1 -3 * -3 -- O

    0 [ F -3 U E + 01 0 1 0 ! "#→#%& (!) ! + "#→#%& (!) "#%&→# ! + "#→#%& (!) 1 F 3 =
  8. . 2 5 5 C 5 33 Tcf ( 1

    g L O 33 2 [] U e h [] N P 33 5 E5 ! − #$ dM - 5 3 %& - 5 3 F 5 5 5 35 2 5 '$  2 5 2 5 '(→$ * Tai ∗ , = 0.4, ∈ = 0.01
  9. . .. S I H C ec . I] 1

    21 . I i  • [ S • OL C H • 1 21 . a H
  10. +14 ( -. . . 1 .- 3. 1 !"

    . . 1 1 4 !# C . . 1 1 4 !# - -
  11. 1 - M NOS C • 2 - 121 2

    - 3 - 2 • 2 - 2 - 3 - I 2 a 3 3. . - !" + !$ !" C 3 . N . M . . C O - 3-4 -
  12. /PI? F? L . L K?L 4?L =K 451 L?E

    T []_ 3 SBLLI K L?E K LM?< ? FI ? ?KMELK ibdhS ? ? ? I L ? /5/ _a /M=E ? KL =? 21 1 % % % 3 SBLLI = E <K ?L L K?LK D LL ? E:K=? ?: IBI <? =BF D ibdhS0E I? =? L ? ? ? MK I P?EK ec gfS/5/ & IP /5/
  13. 1

  14. 1

  15. ?MNM MVKM G H D1 M 2MY VO ]V E

    IWLWVO FIVO VO F] : ] IVL 8IV I] b K VM - 1VV[ NW WX KIT cW ][ VO Xa IU L I X VO IVL KW[ WT]UM 7V 1C ? , G H :71 ?ILPIS [PVI /KPIV I /XX] PIR M V U P /] MT MV :]KKP I[KIT 4]I IJ VM ] e[[ ]VS M IT T K []XM X MT[ KWUXI ML W [ I M WN PM I []XM X MT UM PWL[ 7333 IV[IK WV[ WV XI M V IVITa[ [ IVL UIKP VM V MTT OMVKM () - )d , G(H BV4TW UWV M [ M 8]VP I ] IVL MNIV ?W P BV4TW - BV []XM [ML TMI V VO WN WX KIT cW P I J L MK WVIT KMV[][ TW[[ 7V ///7 M TMIV[ :W] [ IVI , G)H V MT jfhgik 2IV MT 8 0] TM 8WVI[ D]Tl 5I M 0 IVTMa IVL KPIMT 8 0TIKS / VI ] IT [ K WXMV [W] KM UW M NW WX KIT cW M IT]I WV 7V 311C G H 7AA7 jfhgik /VL MI[ 5M OM P T X :MVb IVL ?IY]MT B I[]V / M M MILa NW I] WVWUW][ L VO. PM S [ WV JMVKPUI S [] M 7V 1C ? W b MVbM IVL /VL MI[ 5M OM JRMK [KMVM cW NW I] WVWUW][ MP KTM[ 7V 1C ?