Paper Abstract: Can we make use of data available in the visible domain to develop an algorithm for NIR image segmentation in an Unsupervised domain adaptation scenario? Will bringing targets close to the source to be an effective solution rather than projecting them in domain invariant feature space? Can we make sure that method work for cases where domain shift is more? Since Target data is not always available, can we devise a method such that it is independent of Target data?
Speaker bio:
Aayush is currently working as a Data scientist at Lifebytes. Prior to that, he worked with Dr. Prathosh as a research assistant at IIT-Delhi where he worked to solve skin segmentation using unsupervised domain adaptation. Aayush’s research interest lies mainly in image segmentation, semi/unsupervised learning, and model interpretability. Aayush completed his master's from IIIT-Delhi in Electronics and communication where he worked on pitch disentanglement from speech. He also worked for video analytics and surveillance-based startup on problems like object detection and activity recognition.
Paper link: [ Ссылка ]
