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
Image classification in real world : Car rental damage report
Search
Mathieu Hausherr
September 21, 2018
Technology
0
49
Image classification in real world : Car rental damage report
Mathieu Hausherr
September 21, 2018
Tweet
Share
More Decks by Mathieu Hausherr
See All by Mathieu Hausherr
FrenchKit - Contact Tracing & Exposure Notification
mhausherr
0
110
FRAUG StopCovid19 : Le point sur les applications de tracing Bluetooth
mhausherr
1
230
Swift Lille - CoreML / CreateML
mhausherr
0
64
Mode offline Cocoaheads
mhausherr
0
29
Les surcouches constructeur sur Android
mhausherr
1
740
Mode offline : Notre application Android fonctionne au niveau -5 d’un parking
mhausherr
0
1.2k
Mode offline : Notre application mobile fonctionne au niveau -5 d’un parking
mhausherr
1
110
CocoaHeads : Apple Pay
mhausherr
0
49
Google Glass: Quels Frameworks pour quels usages?
mhausherr
1
84
Other Decks in Technology
See All in Technology
カオナビの利用実績をアウトカムへつなげる旅 / example-of-data-management-startup-in-kaonavi
kaonavi
0
130
Compose Compiler Metricsを使った実践的なコードレビュー
tomorrowkey
1
210
Postman v10リリース後を振り返る / Looking back at Postman v10 after release
yokawasa
1
150
2024/4/26 コンピュータ歴史博物館解説告知
toshi_atsumi
0
210
複雑な構成要素を持つUIとの向き合い方 〜新・支出グラフでの実例〜 / B43 TECH TALK
nakamuuu
0
140
SIEMを用いて、セキュリティログ分析の可視化と分析を実現し、PDCAサイクルを回してみた
coconala_engineer
0
270
4年前、あるじゃん老害エンジニアLT合戦に登壇、米国西海岸コンピュータ歴史博物館体験記の続編
toshi_atsumi
0
220
JAWS-UG Bedrock Claude Night
yamahiro
3
520
よく聞くけど使ったことないソフトウェアNo.1 KafkaとSnowflake
foursue
4
330
継続的な改善 x ⾮連続的な進化
sansantech
PRO
3
130
ここが嬉しいABAC ここが辛いよABAC #再解説+補足編
masahirokawahara
1
270
Reducing Cross-Zone Egress at Spotify with Custom gRPC Load Balancing Recap
koh_naga
0
190
Featured
See All Featured
Adopting Sorbet at Scale
ufuk
68
8.6k
Practical Orchestrator
shlominoach
182
9.7k
Become a Pro
speakerdeck
PRO
11
4.5k
Building Your Own Lightsaber
phodgson
99
5.7k
How To Stay Up To Date on Web Technology
chriscoyier
782
250k
The Pragmatic Product Professional
lauravandoore
25
5.8k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
25
2.3k
Being A Developer After 40
akosma
57
580k
Web development in the modern age
philhawksworth
202
10k
Design by the Numbers
sachag
274
18k
The Invisible Side of Design
smashingmag
294
49k
Teambox: Starting and Learning
jrom
128
8.4k
Transcript
Image classification in real world Car rental damage report
None
None
None
TOP 7 Amazing Tips About Machine Learning for Dummies (aka
iPhone developers)
None
Image classification Folders train Unkown image Classes
#1 ML is not only for data-scientists
None
None
My Dataset 41 269 pictures 62,63 Go Same cars Same
set of colors Same luminosity (no light) Already classified by users
None
None
#2 You already have a dataset or you should
UX proposition Is it a Front alloy wheel ? Yes
sure ! No it’s not
Create ML Top 1 - From 1 34 %
Wheel Tire Fender Gas trap
Create ML Top 1 - From 1 34 % Top
1 - From 5 91 %
UX proposition v2 Something else
Wheel Roof Fender Gas trap
Create ML Top 1 - From 1 34 % Top
1 - From 5 91 % Top n - From n 78 %
#3 Don’t trust results Use it as a hint
None
None
None
Base Merge Classes Top 1 - From 1 34 %
37 % Top 1 - From 5 91 % 93 % Top n - From n 78 % 81 %
#4 Merge classes If you can’t see it, ML can’t
see it
Partner app
Base Partner data Top 1 - From 1 37 %
34 % Top 1 - From 5 93 % 89 % Top n - From n 81 % 75 %
Base Partner data Partner corrected data Top 1 - From
1 37 % 34 % 41 % Top 1 - From 5 93 % 89 % 94 % Top n - From n 81 % 75 % 84 %
#5 Clean your data Size of dataset don’t always matter
None
Inception v3 - MobileNet Inception v3 MobileNet Create ML Top
1 - From 5 91 % 90 % 91 % Top n - From n 79 % 18 % 78 % Size 87 Mo 2.2 Mo 1.5 Mo Training time 2h41 2h38 1h50
#6 Use Create ML It’s really cool
#1 ML is for mobile developer too #2 You already
have a dataset #3 Don’t trust the result #4 If you can’t see it, ML can’t see it #5 Clean your data #6 Use Create ML #7 Drive safe
#1 ML is for mobile developer too #2 You already
have a dataset #3 Don’t trust the result #4 If you can’t see it, ML can’t see it #5 Clean your data #6 Use Create ML #7 Drive safe
Mathieu Hausherr @mhausherr