Image Similarity Detection with TensorFlow

Question

Image Similarity Detection with TensorFlow

Recently I started playing with tensor flow, trying to learn popular algorithms, I am in a situation where I need to find similarities between images.

Image A is fed into my system, and userx supplies image B, and the system should retrieve image A in userx if image B is similar (color and class).

Now I have a few questions:

Are we considering this scenario to monitor learning? I ask because I do not see it as a classification problem (confused !!)
What algorithms should I use for training, etc.
Repeated training should be carried out quite often, how should I deal with this problem, so I do not train every time from scratch (fine tuning?)

+4

scikit-learn deep-learning machine-learning computer-vision tensorflow

iNDicator Nov 17 '16 at 18:49

source share

2 answers

Martin Thoma · Answer 1 · 2016-11-18T15:20:05+0000

Are we considering this scenario to monitor learning?

This is supervised by training when you have tags to optimize your model. Therefore, for most neural networks it is controlled.

However, you can also look at the full task. I think you do not have any truth for pairs of images and the “desired” similarity value that your model should output?

, , CNN ( ), ( ) 1000 . , . , , , .