The candidate will have to create an APP
that allows object recognition and documents classification
(clustering , similarity, topic extraction, etc.)
from from the device-camera live pictures, from taken pictures,
or from local pictures (or scanned documents like PDF containing scans)
by accessing the local device folders.
They will be tickets, invoices, physical products, labels with lot-numbers, labels with best-before-date, etc.
The APP will mostly be used on mobile devices,
but has to be multiplatform, designed in Python by using QT Framework + Felgo libraries.
The APP has to allow to select and use existing public Machine Learning trained databases,
and extend the selected databases with the data created by processing new pictures
(via device-camera, local files and online files via direct URL).
The APP will then allow to extract data from the identified documents,
and fill-in a custom database with the extracted data by creating a digital version of the scanned document,
with the possibility to export the extracted data in XML format too.
Let's imagine an inventory staff man needs to
extract data from a label attached on a product
containing information like a "product lot-number" ad a "best-before-date",
and then digitalize and associate that data to a specific product-name and to
the scanned invoice that lists it.
This is what the staff man should have the possibility to do with the app (in random order):
1) the staff man focuses with the device-camera the invoice paper (listing the items that after
will be focused too), that he will have to mark as arrived to the warehouse:
the focused items that the APP can recognize, will be marked by squares
with their name right below, and the user has to have the possibility
to press 2 buttons on the bottom of the square:
So, an invoice would be marked by a square, with written below:
"Invoice by merchant-name"
The staff man can press one of 2 buttons on the bottom of the square:
OK - "Proceed with this recognition";
ED - "Edit this recognition" ;
when pressing on ED ("Edit this recognition")
the staff man can insert accurate data for the model
to be better trained for future recognition,
marking for example the object as recipe instead of invoice,
or marking as invoice by a merchant instead of another merchant;
when pressing on OK ("Proceed with this recognition")
the APP can classify the kind of document as invoice,
and recognizes the text inside the document,
converting it digitally, and adding its content to the internal
database of the scanned invoices,
associating that particular invoice to the Seller ID
recognized on the invoice itself
(updating the corresponding data if already in the database,
or creating a brand new data if missing).
2) the staff man then focus a physical item in the stock
(a single item, or a pallet, or a package containing multiple items inside,
that will be recognized via the APP with AI thanks to
a Machine Learning model trained on various pictures),
so that the APP recognizes the item(s) focused by the camera,
and simply marks / checks-in the item(s) in the (earlier) digitalized document ;
3) the staff man can then (if needed) make click on the line(s) in the digitalized invoice
containing the information for the recognized item he focused before, so that in a
popup window he can edit the relative quantity of the marked item(s),
if for example to the wharehouse arrive less items than the ones expected.
from the same popup window, the staff man can select a button to take a picture,
so that he can take a picture to the "lot number + best-before-date"
to associate it to the item(s) added / marked as arrived In the digitalized invoice.
There are a lot of ready-to-use open source projects doing one or more things requested for this project.:
I can provide collected links to the freelancer / team that will be hired.
9 freelancers are bidding on average €609 for this job
Buongiorno! I'd like to deliver stock inventory app. I'm familiar with theory of probability and computer graphics. I'll do the job blazingly fast. Please, give me a try!