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

Building a Real-World Pipeline for Image-Classi...

Building a Real-World Pipeline for Image-Classification

How we've build image-classification real-time at heycar. From building the model, up to creating the architecture.

This is the short version, high-overview.

Avatar for Marcelo Boeira

Marcelo Boeira

November 14, 2018
Tweet

More Decks by Marcelo Boeira

Other Decks in Programming

Transcript

  1. ?

  2. ~2 years for 1 person that tags 6 images per

    minute impacts too much on time-to-market
  3. 840 341 269 873 289 269 192 291 465 269

    910 748 289 269 840 341 car text content
  4. 840 341 269 873 289 269 192 291 465 269

    910 748 289 269 840 341 car text content ?
  5. 840 341 269 873 289 269 192 291 465 269

    910 748 289 269 840 341 car text content Tensor Flow
  6. file-storage image-stream Processing image-classifcation-worker Tensor Flow 840 341 269 873

    289 269 192 291 465 269 910 748 289 269 840 341 car: true text_content: false
  7. file-storage image-stream Processing image-classifcation-worker Tensor Flow 840 341 269 873

    289 269 192 291 465 269 910 748 289 269 840 341 image-classification-database car: true text_content: false