4 ms·

Train your own image classifier with Inception in TensorFlow

121 points by rey12rey 11y ago

jszymborski 11y ago
This is great! I was just writing a small script to prepare data for TensorFlow CNN image classification based on a custom dataset using SciKitFlow, but the InceptionV3 model is super cool and it looks like the have an implementation with almost compatible API to what I was writing [1]. I'm super impressed by what's coming out of Google's TensorFlow. Their ImageNet InceptionV3 model is a delight to play with in python! [1] https://github.com/tensorflow/models/blob/master/inception/data/build_image_data.py https://github.com/tensorflow/models/blob/master/inception/d...
- aub3bhat 11y ago
  You might be interested in this, it uses same model to create a very simple bare bones implementation of Approximate Visual Search. https://github.com/AKSHAYUBHAT/VisualSearchServer https://github.com/AKSHAYUBHAT/VisualSearchServer
aub3bhat 11y ago
This is a great news, the earlier released model had some limitations such as it could not be used with a batch. With this and multi GPU training, TensorFlow is now a good alternative to Caffe. Here is my project with TensorFlow inception model. https://github.com/AKSHAYUBHAT/VisualSearchServer https://github.com/AKSHAYUBHAT/VisualSearchServer
- trtg 11y ago
  Is the ami mentioned in your README still available? Searching for ami-b80f0ad2 in community AMIs is not turning up any results.
  - aub3bhat 11y ago
    Yes its available in the Northern Virginia region. https://www.dropbox.com/s/owp7fxwwtlutw8n/Screenshot%202016-03-10%2002.51.11.png?dl=0 https://www.dropbox.com/s/owp7fxwwtlutw8n/Screenshot%202016-...
masonhipp 11y ago
This is awesome --> "In order to make research progress faster, we are additionally supplying a new version of a pre-trained Inception-v3 model that is ready to be fine-tuned or adapted to a new task. We demonstrate how to use this model for transfer learning on a simple flower classification task." Fine-tuning these models for different applications has been a great way for me to build out new things without relying on an enormous fleet of K40s to train a new set from scratch. Lots of progress in this field, thanks to the whole team for releasing this.
ganeshkrishnan 11y ago
Is anyone using Java for running the TensorFlow examples?

121 points by rey12rey 11y ago

This is great! I was just writing a small script to prepare data for TensorFlow CNN image classification based on a custom dataset using SciKitFlow, but the InceptionV3 model is super cool and it looks like the have an implementation with almost compatible API to what I was writing [1]. I'm super impressed by what's coming out of Google's TensorFlow. Their ImageNet InceptionV3 model is a delight to play with in python! [1] https://github.com/tensorflow/models/blob/master/inception/data/build_image_data.py https://github.com/tensorflow/models/blob/master/inception/d...

aub3bhat 11y ago
You might be interested in this, it uses same model to create a very simple bare bones implementation of Approximate Visual Search. https://github.com/AKSHAYUBHAT/VisualSearchServer https://github.com/AKSHAYUBHAT/VisualSearchServer

aub3bhat 11y ago

This is a great news, the earlier released model had some limitations such as it could not be used with a batch. With this and multi GPU training, TensorFlow is now a good alternative to Caffe. Here is my project with TensorFlow inception model. https://github.com/AKSHAYUBHAT/VisualSearchServer https://github.com/AKSHAYUBHAT/VisualSearchServer

trtg 11y ago
Is the ami mentioned in your README still available? Searching for ami-b80f0ad2 in community AMIs is not turning up any results.
- aub3bhat 11y ago
  Yes its available in the Northern Virginia region. https://www.dropbox.com/s/owp7fxwwtlutw8n/Screenshot%202016-03-10%2002.51.11.png?dl=0 https://www.dropbox.com/s/owp7fxwwtlutw8n/Screenshot%202016-...

masonhipp 11y ago

This is awesome --> "In order to make research progress faster, we are additionally supplying a new version of a pre-trained Inception-v3 model that is ready to be fine-tuned or adapted to a new task. We demonstrate how to use this model for transfer learning on a simple flower classification task." Fine-tuning these models for different applications has been a great way for me to build out new things without relying on an enormous fleet of K40s to train a new set from scratch. Lots of progress in this field, thanks to the whole team for releasing this.

ganeshkrishnan 11y ago

Is anyone using Java for running the TensorFlow examples?