BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:Fully Convolutional Pixelwise Context-Adaptive Denoiser
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20261011T174249Z
UID:tag:localist.com\,2008:EventInstance_33547749785700
DTSTART:20200522T150000Z
DTEND:20200522T160000Z
DESCRIPTION:Denoising is a classical problem in signal processing and compu
 ter vision\, and various different methods have been applied to tackle the
  problem for several decades. Recently\, supervised-trained neural network
 -based methods have achieved impressive denoising performances\, significa
 ntly surpassing those of the classical approaches\, such as prior- or opti
 mization-based denoisers. However\, there are two drawbacks on those metho
 ds\; they are not adaptive\, i.e.\, the neural-network cannot correct itse
 lf when distributional mismatch between training and test data exists\, an
 d they require clean source data and exact noise model for training\, whic
 h is not always possible in some practical scenarios. In this talk\, I wil
 l introduce a framework that tries to tackle above two drawbacks jointly\,
  based on an unbiased estimate of the loss of a particular class of pixelw
 ise context-adaptive denoisers. Using the framework and neural networks to
  learn the denoisers\, I show the resulting image denoiser can adapt to mi
 smatched distributions in the data solely based on the given noisy images\
 , and achieve the state-of-the-art performances on several benchmark datas
 ets. Moreover\, combined with the standard noise transform/estimation tech
 niques\, I will show that our denoiser can be completely blindly trained o
 nly with the noisy images (and without exact noise model) and yet be very 
 effective for denoising more sophisticated\, source-dependent real-world n
 oise\, e.g.\, Poisson-Gaussian noise.\n\nAt the end of the talk\, if time 
 permits\, I will also briefly talk about my other projects focusing on ada
 ptivity and stability of machine learning methods\, e.g.\, on continual le
 arning and explainable AI methods.
LOCATION:Zoom
SUMMARY:Fully Convolutional Pixelwise Context-Adaptive Denoiser
URL;VALUE=URI:https://events.seas.harvard.edu/event/fully_convolutional_pix
 elwise_context-adaptive_denoiser
CATEGORIES:Colloquia / Seminar / Lecture
END:VEVENT
END:VCALENDAR