[DLHacks]Random Erasing Data Augmentationとデータの前処理module@tensorflow

6 21 5/ 0

• R • R – D E A D

d • d ce b Si b sp – 2 C :HF H DC h l – 3 F E A 2 C :HF H DC pu Mx c l – 4 C D -FDE rotZ cp [ – 1DF DCH A 0A E d e m ibP – A , CH H DC l c rotZ p [ – , E H 4 HD , CH H DC dns p l7( ) ) ( – -DADF , CH H DC CC Al( d p R3-, – 4 C D • • • dg a /F C e DC z cb e c C A l e A 4 C D c /F C . H , h CH H DC ez 4 C D d /F ] C b

• https://arxiv.org/abs/1708.04896

5: GE 1 : a T • 5: GE 1 : f Z Te Wcd v T – 9A 9AG D :D 5: GE 1 : 0: : . E : G 5: 1: bw Rgkqip Wtuns – 6 : 0 7 2 :A:E 8 6: DG 3E GN 5 D: : G G /G NGD G :D :D GC A/ G / G hl • o m e atuns z frki e T • itk R yR A I : GE aVT : G E b , f ( ) GE

• https://arxiv.org/abs/1708.04896
• https://arxiv.org/abs/1708.04552
• https://qiita.com/dsanno/items/d32f3c928cdbcbe5de60

• ER T E N R E a

: 3:3 @ / 2 : ) / 7 • • Eb – @@ – L l 7@ 1 : Tl o T fI w 3 i hg _d aw mu hg e_ c 71 7 ( 3 pk e_ hg rL nt Rs 13 .7:/ 3 y :/ @3 3 13

• https://github.com/koichiro11/preprocess_image/blob/master/preprocessor.py

• • : 0 1 1 0 . 1 0 / 0

• https://github.com/koichiro11/preprocess_image/blob/master/data_augmentation_test.ipynb

2 1 ) • 2 1 2 1 M a – • u d F 0 B E E2 1 i R L ( BB? ( ? B B T Mt cM C77 Lnemsof p l F i ghld P c w d L C /. : F 7 B7 B wM M rs a rs

• https://deepage.net/tensorflow/2017/10/07/tfrecord.html

njlt i – :EEAD • .11 N – :EEA • 23 6@ – :EEAD • a b qrgh D 6 D:3C D 6 D:3C E C DD mucs epf dpf :3E 36 3C A EC 0C3D 9 /3E3 - 9 E3E @ o E3 5@ 6D3 @ E D 6 8 5 56454 6 R

• https://www.slideshare.net/ren4yu/ss-84282514
• http://iwiwi.hatenadiary.jp/entry/2016/12/31/162059
• https://qiita.com/dsanno/items/d32f3c928cdbcbe5de60