If you look at the flower in front of you, you will discover its name.
The app is currently able to recognize among 1000 species of flowers.
It does not have the ambition to replace the eye of a botanical expert, but, for the flower that you are at that time framing with your smartphone, offers up to three possible species names.
It is only a suggestion, a first step for further necessary verification.
The app has experimental value and aims to test the power of convoluted neural networks, which are the new frontier of deep learning.
The open source TensorFlow framework and the Inception V.3 model are used, which exceeds version 1, winner of the 2014 Imagery Large Visual Recognition Challenge, with a margin of error that goes from 6.67% of version 1 to 3.46%.
The model is specially retrained for the recognition of flowers.
Thanks to Oracle for making the necessary resources available.
A few years ago, when my son was young, we took him on a trip to an historic villa near Rome. Looking at the trout that swam in one of the pools of the villa, he asked me if there was an app that could recognize the animal species that was framed with the smartphone. I replied that it did not exist, but that we could do it together. At home we threw down some sketches on a piece of paper and then it was forgotten in a drawer. The business was above my chances, but I always had the idea that if I had found an idea I would have taken it up again.
Suddenly the thing was reborn because I discovered that someone else, with knowledge and technology far more advanced than mine, had brought it forward: the TensorFlow project, the neural network of Google called Inception-v3 and the mobile porting through the TFMobile framework.
'1000 Flowers' is a first initial step to see realized that far, and by now forgotten, my son's desire: the recognition of most of the living species.
The app is currently able to recognize among 1000 species of flowers.
It does not have the ambition to replace the eye of a botanical expert, but, for the flower that you are at that time framing with your smartphone, offers up to three possible species names.
It is only a suggestion, a first step for further necessary verification.
The app has experimental value and aims to test the power of convoluted neural networks, which are the new frontier of deep learning.
The open source TensorFlow framework and the Inception V.3 model are used, which exceeds version 1, winner of the 2014 Imagery Large Visual Recognition Challenge, with a margin of error that goes from 6.67% of version 1 to 3.46%.
The model is specially retrained for the recognition of flowers.
Thanks to Oracle for making the necessary resources available.
A few years ago, when my son was young, we took him on a trip to an historic villa near Rome. Looking at the trout that swam in one of the pools of the villa, he asked me if there was an app that could recognize the animal species that was framed with the smartphone. I replied that it did not exist, but that we could do it together. At home we threw down some sketches on a piece of paper and then it was forgotten in a drawer. The business was above my chances, but I always had the idea that if I had found an idea I would have taken it up again.
Suddenly the thing was reborn because I discovered that someone else, with knowledge and technology far more advanced than mine, had brought it forward: the TensorFlow project, the neural network of Google called Inception-v3 and the mobile porting through the TFMobile framework.
'1000 Flowers' is a first initial step to see realized that far, and by now forgotten, my son's desire: the recognition of most of the living species.
Show More
