Master’s student

Instance segmentation for defect detection on civil infrastructure

Ref. 2021_061

This project aims to develop an instance segmentation approach mainly based on the Detectron2 library for detecting various defects that can be found on civil infrastructure.

 

Description

Aging civil infrastructure requires continuous monitoring to assess its safety of operation and continued usage, think of bridges, roads, etc. Technologies today such as drone imaging is making it easier to acquire high resolution images of such infrastructures. However, analysing such data is very costly and difficult in many cases due to the massive volumes of data that is acquired. This makes a perfect use case for deep learning approaches which have already established themselves as the go to methods for analysing images. Therefore, techniques such as Object Detection and Instance Segmentation are being used to analyse these high volumes of images and automatically detect defects. Yet many of these techniques still present many shortcomings. Some of these are the need for manually labelling a great number of these images prior to having supervised training which can be very costly and, in many cases, impossible. Also, the incapability of detecting smaller defects which might be of high importance.
Therefore, this thesis should make use of data efficient techniques such as self-learning and semi-supervised learning to leverage the available labelled data and achieve or even surpass supervised learning performance.

 

Goal

The goal of this master thesis is to make use of and improve on state-of-the-art methods for instance segmentation. This should allow to mitigate some of the shortcomings of generic instance segmentation libraries on public datasets such as COCO, LVIS and especially in real world scenarios.

 

Minimum qualifications

  • Strong coding skills for Deep Learning (Python, Pytorch)
  • Experience in one or more of:
    • Object Detection
    • Semantic Segmentation
    • Instance Segmentation
  • Proficient working in Unix/Linux environments

 

Preferred qualifications

  • Scientific approach for conducting experimentation and working on research papers
  • Familiarity with Detectron2 is a strong plus
  • Familiarity with data augmentation libraries such as Albumentations or others is a plus

 

Diversity

IBM is committed to diversity at the workplace. With us you will find an open, multicultural environment. Excellent flexible working arrangements enable all genders to strike the desired balance between their professional development and their personal lives.

 

How to apply

If you are interested in this role, please submit your application below.

 

Illustration