Detection, segmentation, and multiple rules for determining damage
A task
The task of the project was to achieve uniformity of the brand’s gas stations. The customer has thousands of objects and more than 20 brands, so it was necessary to do this without involving third-party people who should inspect the inconsistencies of the object. The idea was that the clients of the objects themselves send the necessary photos, for which they receive a bonus to the loyalty card.
Solution
A prototype was developed and a neural network was trained, ranging from obtaining photos from scratch to the simple web interface in which you can test the operation of a trained neural network.
The necessary photos were collected through crowdsourcing. From several tens of thousands of photos it was necessary to extract the necessary content.
Identify the desired objects in images and add labels to them.
1.
External Data Collection
3.
Neural Network Training
4.
Developing a User Interface
2.
Data Markup
Bisiness effect
About process of work
The project was successful, and an action was held for customers: for the photos they took on their phone, they received a bonus from the company. We managed to create a rather interesting information channel and involve the client in the bonus program.
The case is useful for any companies that want to monitor the status of marketing materials.
The case is useful for any companies that want to monitor the status of marketing materials.
Project team
Frank Sh.
Developer
