Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Probabilistic Integration of Deep Learned Defect Detections into a 3D Model
Visual inspection using drones is playing an increasingly important role in many branches of industry. We want to integrate the output of a neural network, trained to detect these defects into a volumetric representation of the scene.
Visual inspection using drones is playing an increasingly important role in many branches of industry. We want to integrate the output of a neural network, trained to detect defects in visual inspection images into a volumetric representation of the scene. This integration step takes detection scores from the neural network output and fuses them in the volumetric map. When extracting a surface these scores should be projected to the surface and form a surface which indicates defects on the 3D model.
This thesis will be a joint project between ASL and Voliro (www.voliro.ch), an ETH-spin off developing the next generation of drones for aerial inspection and manipulation.
Visual inspection using drones is playing an increasingly important role in many branches of industry. We want to integrate the output of a neural network, trained to detect defects in visual inspection images into a volumetric representation of the scene. This integration step takes detection scores from the neural network output and fuses them in the volumetric map. When extracting a surface these scores should be projected to the surface and form a surface which indicates defects on the 3D model.
This thesis will be a joint project between ASL and Voliro (www.voliro.ch), an ETH-spin off developing the next generation of drones for aerial inspection and manipulation.
- Literature review
- Setting up/finding an appropriate detection pipeline
- Training and evaluating the detection pipeline
- Integration of depth images together with the RGB images and the detection scores
- Creating visualizations and evaluations
- Literature review - Setting up/finding an appropriate detection pipeline - Training and evaluating the detection pipeline - Integration of depth images together with the RGB images and the detection scores - Creating visualizations and evaluations
- Python
- Deep learning
- At least one of the three: Pytorch/Keras/Tensorflow is a plus
- Knowledge about TSDF is a plus
- Python - Deep learning - At least one of the three: Pytorch/Keras/Tensorflow is a plus - Knowledge about TSDF is a plus
Send CV and transcript to:
- Fadri Furrer (fadri.furrer@mavt.ethz.ch)
- Marius Fehr (marius.fehr@mavt.ethz.ch)
Send CV and transcript to:
- Fadri Furrer (fadri.furrer@mavt.ethz.ch) - Marius Fehr (marius.fehr@mavt.ethz.ch)