Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Domain-Adaptive Semantic Segmentation for Robotic Navigation in Natural Environments
Natural environments are inherently dynamic and heterogeneous, characterized by significant variability in appearance and structure. The primary goal of this project is to develop a training scheme capable of adapting lightweight semantic segmentation models to changing environments without relying on manually annotated data. In particular, the student will investigate advanced domain adaptation techniques to bridge the gap between synthetic and real data, as well as between sets of real images collected during different seasons. The developed approach is expected to enhance the model’s generalization capabilities, allowing it to adapt more rapidly to unseen conditions, and ultimately improving the robots' operational efficacy and robustness.
Not specified
-WP1: Literature research on Unsupervised Domain Adaptation (UDA) for Semantic Segmentation.
-WP2: Design and implementation of an UDA pipeline, focusing on lightweight architectures.
-WP3: Experiments with simulated and real data.
-WP4: Evaluation of the proposed approach and comparison against the state of the art.
-WP1: Literature research on Unsupervised Domain Adaptation (UDA) for Semantic Segmentation. -WP2: Design and implementation of an UDA pipeline, focusing on lightweight architectures. -WP3: Experiments with simulated and real data. -WP4: Evaluation of the proposed approach and comparison against the state of the art.
-Background in Computer Vision and Deep Learning.
-Python programming experience.
-Background in Computer Vision and Deep Learning. -Python programming experience.
Please send **CV** and **Transcripts** to Lucas Teixeira, lteixeira@mavt.ethz.ch and rmascaro@ethz.ch
Please send **CV** and **Transcripts** to Lucas Teixeira, lteixeira@mavt.ethz.ch and rmascaro@ethz.ch
