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!
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.
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.
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/d...