Skip to main content

July 2021

In July, we added to the portal the tasks section, the database failover, we improved the shelf detector, and added the price tags on goods presence indicator.

Tasks on the portal

In this update, we have added tasks to the portal. This will come in handy if you need a one-time check of the goods on the shelves during your next point of sale visit.

You can create a task in the Task List section. Also, filters have been added to the section. For example, filters by status, due date or store.

The way it works

Во время визита, мерчендайзер фотографирует товар и отмечает, выполнена задача или нет. Если нет, то указывает причину, по которой задачу нельзя выполнить.

After that, the manager checks the visit results.

All changes to the task are saved in the event history.

Failover Database

Every program has a database, an organized information repository. If errors occur in the database operation, access to the data is blocked. The user cannot write or retrieve information until the database is back in service.

To improve the fault resistance of the database, and eliminate the downtime, we have added the failover. This is an emergency service switch to a backup database, during the main database failure.

The way this works

Each database has a copy: a duplicate database on a separate server, which is synchronized with the main database.

If there are problems with the main database, the system will automatically switch to a copy, and the service will continue to work without data loss.

When the problems are fixed, the data is synchronized with a copy, and the restored database again becomes the main one.

Shelf detector dataset

In March, we talked about the new Neural Network detector of shelves and racks. We continue to train the detector for more accurate recognition.

Changes implemented

Sometimes in the visit scenes there are photos with unusual forms of shelves. For example, potato chips in a store are not sold off the shelf, but rather straight out of the box. This is not a typical situation for the detector, and it needs to be trained to recognize such data.

To do this, we made a special dataset, in which you can mark such options for shelves in a photo. The neural network can use this data for training.

In the near future, we plan to collect information from datasets, train the detector, and keep it up to date.

Recognition of price tags presence

One of the company requirements for the store and its employees, is that all goods at the point of sale must have price tags. In order for a company’s analyst to check this, we have added a separate type of business indicator.

Now, even if a price tag in a photo is blurred or with glare, the system would still recognize its presence, and transfer this indicator along with other data for export.