Panoptic segmentation on point cloud data presents a new challenge in exploiting the merits of both detection and segmentation, with the aim of unifying instance segmentation and semantic segmentation in a single framework. However, an efficient solution for panoptic segmentation in the emerging domain of LiDAR point cloud is still an open research problem and includes many challenging situations. Understanding point cloud data is inherently an unordered list of points and therefore, traditional deep learning approaches can not be applied directly.
SAMP has access to a large amount of data to train machine learning models. However, the scans come from many different sites that may differ in their nature (Nuclear Plants, Oil and Gas plants, Chemical Plants, Manufacturing Plants) as well as the characteristics of the sensors. So data imbalance and redundancy are a systematic characteristic that arises when collecting large amounts of data. Understanding our data distribution helps in choosing the right set of data samples that are important for model training. Furthermore, data augmentation techniques are essential to avoid model overfitting.
You will be working with our automation team on improving our panoptic segmentation pipeline. The objective of this ambitious internship is to scale up the performance of our current AI models for point cloud segmentation. Your tasks will include:
- Understanding and familiarizing with SAMP AI models for panoptic semantic segmentation, as well as our internal datasets
- Developing and evaluating deep learning algorithms on panoptic segmentation of point cloud dataset
- Investigating different data augmentation techniques and analyze their effects on our model performance.
Overall, you are expected to be proactive and contribute with your ideas to improve the results.
- Contract Type: Internship (Between 5 and6 months)
- Start Date: Jan-Feb 2022
- Location: Paris 6ème, France
- Education Level: Master’s Degree or Equivalent
- Salary: 1200€ gross / month
- Advisor: Nachwa Bakr, PhD and Shivani Shah, PhD
- Possible partial remote