Which department will you join?
How hard is it to identify vehicles in a picture?
It is not very difficult to identify a normal car, but on the road we may encounter cases that are not routine.
The challenge our team faces is to correctly identify what meets the definition of a "vehicle" and what does not. The task is challenging since we need to do it within real-time constraints and with great accuracy.
To achieve performance appropriate for a real ride, we use very large amounts of tagged data.Our team's developers face many tasks that include all aspects of work with data:
How do we make sure that the data is clean, varied, and sampled correctly?
How do you take raw data from tagged car ride videos and turn it into examples that go into a neural network?
Once we have good data, we try to get the most out of it. We train a network that generalizes well for cases in the real world
Vehicle identification is critical for any driving-assistance system, so our team is at the core of the company's algorithmic activity. We are at the forefront of development, both in terms of tools and in terms of algorithmic capabilities.
What will your job look like:
Development is done mostly in Python, using numpy and pandas. We use mint, a tensorflow-based framework tailored to our needs, and quite a few internal tools written by the company or by the team. We also work with C++ from time to time, and it is good to have knowledge of the Linux environment.