How to implement bottle recognition by its photo?

Question

Python

There is a fairly large collection of images of bottles of alcohol in good quality. With a textual description of what kind of bottle it is.
There is a desire to make a bottle recognizer based on this collection. For example, in the form of a telegram bot, to which the user sent a photo of the bottle, and in response received a text describing the bottle.
The only problem is that I don’t know where to start and what technologies to use in terms of bottle recognition. What libraries and technologies to use. What is an example procedure. The language will be Python.
I would appreciate all opinions and advice.
Thank you!

Answer 1

bubublik, if I were you, I would take some well-known grid (inception, for example), load imagenet weights. Next, freeze the upper layers and retrain on your dataset. In essence, you will get a regular classifier

Answer 2

Good day. Well ... the start is given. Python is quite powerful in all aspects of your idea.
I propose to proceed as follows:
1) To begin with, train the network in Google Colab. If there is an opportunity on capacities on house PC - it is possible on it. Split your photos into training and test sets and see the results. For all this you will need a CNN (Convolutional Neural Network), I suggest you first familiarize yourself with the concept of object recognition using CNN.
2) Write an API for the final model of the network
3) Write a bot, testing the basic functions that you need (eg uploading photos).
4) Connect the neural network model and the bot using the written API