Build a Mini-Amazon-Go in 1 Day Nov 2019 Zhi Zhang, Rachel Hu - - PowerPoint PPT Presentation

▶

Mar 30, 2024 149 likes •281 views

Build a Mini-Amazon-Go in 1 Day Nov 2019 Zhi Zhang, Rachel Hu Amazon AI d2l.ai Outline Initiate the work-horse(SageMaker notebook) Showcase a demo from the noob Sampling real-world data Hello world for object labeling

SLIDE 1

d2l.ai

Zhi Zhang, Rachel Hu Amazon AI

Build a Mini-Amazon-Go in 1 Day

Nov 2019

SLIDE 2

numpy.d2l.ai/gluon-cv.mxnet.io

Outline

Initiate the work-horse(SageMaker notebook)
Showcase a demo from the noob
Sampling real-world data
Hello world for object labeling
Train and Evaluate object detector
Happy hacking!
Leaderboard

SLIDE 3

numpy.d2l.ai/gluon-cv.mxnet.io

Initiate the work-horse

*@request-nb.mxnet.io

SLIDE 4

numpy.d2l.ai/gluon-cv.mxnet.io

Showcase a demo from the noob

+ <= $60

SLIDE 5

numpy.d2l.ai/gluon-cv.mxnet.io

Showcase a demo from the noob

SLIDE 6

numpy.d2l.ai/gluon-cv.mxnet.io

Showcase a demo from the noob

SLIDE 7

numpy.d2l.ai/gluon-cv.mxnet.io

Sampling real-world data

SLIDE 8

numpy.d2l.ai/gluon-cv.mxnet.io

Hello world for object labeling

https://www.makesense.ai/

SLIDE 9

numpy.d2l.ai/gluon-cv.mxnet.io

Train and Evaluate object detector

SLIDE 10

numpy.d2l.ai/gluon-cv.mxnet.io

Happy Hacking

Team up
Split the tasks
Sampling training images
Parallel labeling
Train and evaluate object detection models
Design your cashier-less logic

SLIDE 11

numpy.d2l.ai/gluon-cv.mxnet.io

Evaluation Method

Accuracy: percentage of correctly detected transaction

movements

UX(User experience): how satisfying the overall user

experience is.

Innovation: how you bring cool technologies into the

pipeline

SLIDE 12

numpy.d2l.ai/gluon-cv.mxnet.io

Troubleshooting

GPU: out of memory
If there are too many notebooks opened which occupies

lot’s of GPU memory, then you should close some of them to avoid OOM error