Internship in computer vision dataset development and CNN model

Company Description

Convinced that technology should contribute to making our living environments simpler, sustainable and safer, LACROIX (5300 employees, €619M pro-forma turnover) supports its clients in the design and management of smart living ecosystems, thanks to connected equipment and technologies and through a vision:

"Connected Technologies for a Smarter World".

At the heart of the Group's L25 strategy, Impulse is the R&D Business Unit of our Electronics activity. It brings together key technologies to support companies and industries in accelerating their technological and digital transformation. Impulse's offer is based on a complete range of expertise in designing and industrializing IoT solutions (hardware, software and cloud), AI, Computer Vision, Cybersecurity, and more to become a leading player in industrial IoT. Impulse targets high stake societal markets with growth perspectives, particularly in the fields of home automation, industry, mobility and automotive.

Mainly based in Cesson in the heart of the Rennes technological basin, the Impulse teams (currently 100 employees) are set to grow with LACROIX by contributing to its growth.

Job Description

Subject: Dataset and CNN model for road object detection

Detection and tracking of objects in videos offer a wide range of applications in robotics and in our daily life. To do this, the use of deep neural networks has become essential due to their great capacity to learn meaningful visual representations. At LACROIX-Impulse, the LACROIX R&D center based in Rennes Metropolis, we have considered road object detection as a goal for traffic surveillance and regulation.

You will be attached to Lesley-Ann, a Computer Vision Engineer.

The objective of the internship is to study the use of data augmentation, such as synthetic data generated by the CAR Learning to Act (Carla) simulator, in order to improve object detection under certain conditions. This simulator allows to generate training data (images, optical flow, segmentation, trajectories...) on a large scale with specific objects and parameters encountered in real-world video scenes, but less available in existing datasets like COCO or Google OpenImage. The use of the AirSim simulator may also be considered as well as the creation of a virtual city by the intern. Improvements on the existing network may also be proposed.

Your tasks:

• Learn to use the CARLA simulator to generate synthetic training data with the considered objects and parameters.
• Create a user interface to automate the creation of these new datasets.
• Compare the CARLA simulator with its competitor AirSim.
• Perform processing, fusion, and post-processing of the data in order to prepare the training and test data for the network.
• Set up metrics to analyze the datasets (existing and synthetic).
• Train the CNN based object detection model with the augmented dataset.
• Set up benchmarking to quantify the impact of the addition of new datasets on the network's performance depending on their characteristics.
• Optional: work on creating a new map usable on CARLA by mapping a district of Rennes with OpenStreetMap.
• Write a report, a scientific article or an invention disclosure depending on the results obtained.

Expected results

• A user-friendly interface allowing the generation of synthetic data from a simulator for the task,
• An augmented dataset for training and testing the CNN based road object detection model which is a fusion of the synthetic and existing datasets,
• The training and testing of the CNN model for the task, with an evaluation to confirm the improvement in detection performance,
• Metrics for analyzing the entire dataset (synthetic and existing)
• A report and a patent application depending on the novelty of the approach studied.

Qualifications

• You are a student in engineering school or equivalent
• People say you are creative and autonomous.
• You have a certain passion for algorithms, especially in computer vision and you have good Python programming skills.
• Successful experience in a machine learning framework and experience in using 3D engines (Unity, Unreal Engine) would be a plus (Pytorch, TensorFlow, etc.).
• Knowledge in C, C++ or C# would also be a plus.
• You may need to work in an international context, so a good level of English is required.

Additional Information

Benefits:

• Paid internship with access to restaurant vouchers
• Opportunity to work on innovative and high-tech projects
• Supervision by an experienced and passionate team