I'm writing a mobile app that will enable a user to scan a Craft Beer Label from a bottle, tap, six pack, etc.

The scan will only work for my customers who are the Brewers themselves, so I will have access to all of the artwork used for their labels.

My concern is around training the model. As you can imagine, there will be differences in lighting conditions at bars, amoung other challenges.

As I am just getting started in Machine Learning, is this task feasible? How difficult will it be to train the model given the various conditions found at different pubs/bars.



For sure, is a really feasible application! Indeed, there does exist Deep Learning models for that.

I suggest you to start by doing transfer learning on a pre-trained model like this: https://github.com/satojkovic/DeepLogo Typically you remove the last layers of the network and add layers that adapt to your data. Then, train these last new layers with your data (freezing the layers you have kept from the original model).

Another good advice is to perform data augmentantion on your dataset to include ligth, scale and rotation variants of your images.


  • It's a feasible task
  • Check out pre-build models
  • Learn about transfer learning on pre-trained models
  • Learn about data augmentation to include robustness against different scenarios and env. conditions
| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.