BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Detecting dangers in traffic scenes
DTSTART:20180626T133000
DTEND:20180626T153000
DTSTAMP:20260406T103744Z
UID:9d6da117f43a718095c617896b0a4ab0f9972f41d693cd14fcf902fc
CATEGORIES:Conferences - Seminars
DESCRIPTION:Krzysztof Lis\nEDIC candidacy exam\nExam president: Dr. Martin
  Rajman\nThesis advisor: Prof. Pascal Fua\nThesis co-advisor: Dr. Mathieu 
 Salzmann\nCo-examiner: Dr. Micahël Thémans\n\nAbstract\nSelf-driving car
 s will have to flexibly react to unpredictable dangers.\nHowever many mode
 rn computer vision techniques rely on learning from datasets containing on
 ly typical situations.\nOur goal is to detect unusual dangers (animals\, r
 are vehicles and obstacles\, anything not present in the datasets) in imag
 es of traffic scenes.\nWe discuss three articles related to our task:\nan 
 instance-aware semantic segmentation method InstanceCut - the base model o
 f our scene understanding\;\na class-agnostic object proposal generator De
 epMask\, capable of generalizing to previously unseen object categories\;\
 nand a conditional GAN called pix2pixHD which generates images from semant
 ic labels\, possibly allowing us to detect errors in semantic segmentation
  in an unsupervised way.\nWe present our results and plans for further res
 earch.\n\nBackground papers\nInstanceCut: from Edges to Instances with Mul
 tiCut\, by Kirillov\, A.\, et al.\nLearning to Segment Object Candidates\,
  by Pinheiro\, P.\, et al.\nHigh-Resolution Image Synthesis and Semantic M
 anipulation with Conditional GANs\, by Wang\, t-C.\, et al.\n\n 
LOCATION:BC 229 https://plan.epfl.ch/?room==BC%20229
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR