BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Monocular Depth Estimation via Style Transfer
DTSTART:20200226T130000
DTEND:20200226T150000
DTSTAMP:20260501T144335Z
UID:adc4975dd5bd07921879ce6fb73466129f3b61e9980a8b2e0f3f9c13
CATEGORIES:Conferences - Seminars
DESCRIPTION:Deblina Bhattacharjee\nEDIC candidacy exam\nExam president: Pr
 of  Robert West\nThesis Advisor: Prof Pascal Fua\nThesis co-advisor: Dr. 
 Mathieu Salzmann\nCo-examiner: Prof. Pierre Dillenbourg\n\nAbstract\nEstim
 ating depth for comic images is more aptly a monocular depth estimation pr
 oblem and is subject to the availability of ground truth data while using 
 learning-based approaches. However\, most monocular depth estimators eithe
 r need to rely on large amounts of ground truth depth data\, which is extr
 emely expensive and difficult to obtain\, or predict disparity as an inter
 mediary step using a secondary supervisory signal leading to blurring and 
 other artefacts. Training a learning-based depth estimation model using sy
 nthetic pseudo ground truth data can resolve some of these issues but intr
 oduces domain bias. To remedy the problem of domain bias\, in this report\
 , we have applied style transfer and adversarial learning via GANs. We als
 o apply a depth estimator simultaneously on the style-transferred images t
 o reduce memory footprint and time overhead. Therefore\, our model estimat
 es pixel-level depth from a single comics image based on its training over
  a large amount of readily available real-world data and its corresponding
  depth ground truth. The experimental results and qualitative analysis sho
 w the achieved accuracy and robustness of this method when compared to pre
 -existing techniques. Whilst the results have scope for improvement\, our 
 preliminary study shows that this approach can lead to an accurate pixel-l
 evel depth estimation.\n\nBackground papers\nTowards Instance level Image-
 to-ImageTranslation\, by Shen\, Z\, et al.\nCross-Domain Weakly Supervised
  Object Detection through Progressive Domain Adaptation\, by Inoue\, N\, e
 t al.\nTowards Robust Monocular Depth Estimation: Mixing Datasets for Zero
 -shot Cross-dataset Transfer\, by Ranftl\, R.\, et al.
LOCATION:ELD 120 https://plan.epfl.ch/?room==ELD%20120
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR