CLASSIC VS DEEP VISION:
WHAT IS BEYOND DEEP LEARNING IN COMPUTER VISION?
THE SINGLE IMAGE ARTEFACTS REMOVAL CASE
TUTORIAL @ACCV18, 3RD DECEMBER - PERTH, AUSTRALIA
TUTORIAL @ACCV18, 3RD DECEMBER - PERTH, AUSTRALIA
LOCATION: Level 2 Riverview, Room 5 (Map HERE) TIME: 9AM
LOCATION: Level 2 Riverview, Room 5 (Map HERE) TIME: 9AM
ORGANISING TEAM
ORGANISING TEAM
UNIVERSITY OF CAMBRIDGE
YALE-NUS COLLEGE
UNIVERSITY OF CAMBRIDGE
DATA61/ ANU
SHANDONG UNIVERSITY
THE GEORGE WASHINGTON UNIVERSITY
OVERVIEW
OVERVIEW
The advent of Deep Learning (DL) in computer vision – since the pioneering work of Hinton [1] in 2012 – changed the perspective of the community, adopting in this way DL as the go-to technique for different computer vision tasks. This emergence is justified by DL’s superior performance on various vision tasks including classification, recognition and image segmentation. However, despite the fact that DL is a powerful tool, there is still a missing gap in understanding its properties, which is not the case with more classical computer vision approaches as they are more tractable, and often offers a clear understanding about how they work.
The advent of Deep Learning (DL) in computer vision – since the pioneering work of Hinton [1] in 2012 – changed the perspective of the community, adopting in this way DL as the go-to technique for different computer vision tasks. This emergence is justified by DL’s superior performance on various vision tasks including classification, recognition and image segmentation. However, despite the fact that DL is a powerful tool, there is still a missing gap in understanding its properties, which is not the case with more classical computer vision approaches as they are more tractable, and often offers a clear understanding about how they work.
In this tutorial, we shall draw attention to a new direction that merges the mathematical benefits of classical vision and the powerful performance of deep learning. We will start by introducing the topic and giving an overview of both classic and deep vision perspectives. We will then present a case study of the single image reflection removal problem in which we share solutions coming from the two perspectives. This shall be followed by a discussion of how failure cases - in both perspectives - are related to the modelling hypothesis, and how these failures motivate the need for combined both perspectives. This shall motivate the question - What Is Beyond Deep Learning In Computer Vision? Some open questions related to the topic will also be discussed in the end.
In this tutorial, we shall draw attention to a new direction that merges the mathematical benefits of classical vision and the powerful performance of deep learning. We will start by introducing the topic and giving an overview of both classic and deep vision perspectives. We will then present a case study of the single image reflection removal problem in which we share solutions coming from the two perspectives. This shall be followed by a discussion of how failure cases - in both perspectives - are related to the modelling hypothesis, and how these failures motivate the need for combined both perspectives. This shall motivate the question - What Is Beyond Deep Learning In Computer Vision? Some open questions related to the topic will also be discussed in the end.
SCHEDULE
SCHEDULE
PART 1: Classic vs Deep Learning Vision – An Overview
PART 1: Classic vs Deep Learning Vision – An Overview
1a. Tutorial Overview - Carola-Bibiane Schönlieb
1a. Tutorial Overview - Carola-Bibiane Schönlieb
1b. Classic vs Deep Vision - An Overview - Shaodi You
1b. Classic vs Deep Vision - An Overview - Shaodi You
------- BREAK (15 MIN)
------- BREAK (15 MIN)
PART 2: Classic vs Deep Learning Vision (The Single Image Artefact Removal Case)
PART 2: Classic vs Deep Learning Vision (The Single Image Artefact Removal Case)
2a. Revising the Classic Perspective - Angelica I. Aviles-Rivero
2a. Revising the Classic Perspective - Angelica I. Aviles-Rivero
2b. Revising the Deep Learning Perspective - Qingnan Fan
2b. Revising the Deep Learning Perspective - Qingnan Fan
------- BREAK (15 MIN)
------- BREAK (15 MIN)
PART 3: What is Beyond Deep Learning in Computer Vision?
PART 3: What is Beyond Deep Learning in Computer Vision?
3a. What is Beyond Deep Learning in Computer Vision? - Robby T Tan
3a. What is Beyond Deep Learning in Computer Vision? - Robby T Tan
Panel Discussion
Panel Discussion
MATERIALS
MATERIALS
Tutorial Flyer [Download]
Tutorial Flyer [Download]
PART 1a. C.B. Schönlieb [Download]
PART 1a. C.B. Schönlieb [Download]
PART 1b. S. You [Download]
PART 1b. S. You [Download]
PART 2a. A.I. Aviles-Rivero [Download]
PART 2a. A.I. Aviles-Rivero [Download]
PART 2b. Q. Fan [Download]
PART 2b. Q. Fan [Download]
PART 3. R.T. Tan [Download]
PART 3. R.T. Tan [Download]
REFERENCES
REFERENCES
- Tan, R. T. Visibility in bad weather from a single image. IEEE Conference in Computer Vision and Pattern Recognition (CVPR), 2008.
- Krizhevsky, A., Sutskever, I., & Hinton, G. E. Imagenet classification with deep convolutional neural networks. In Advances In Neural Information Processing Systems (NIPS) (pp. 1097-1105), 2012.
- Li, Y., Brown, M.S.: Single image layer separation using relative smoothness. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2752–2759, 2014.
- Li, Y., You, S., Brown, M. S., & Tan, R. T. Haze visibility enhancement: A survey and quantitative benchmarking. Computer Vision and Image Understanding, 165, 1-16, 2017.
- Fan, Q., Yang, J., Hua, G., Chen, B., & Wipf, D. A generic deep architecture for single image reflection removal and image smoothing. IEEE International Conference on Computer Vision (ICCV), 2017.
- Heydecker, D., Maierhofer, G., Aviles-Rivero, A. I., Fan, Q., Schönlieb, C. B., & Süsstrunk, S. Mirror, Mirror, on the Wall, Who's Got the Clearest Image of Them All? - A Tailored Approach to Single Image Reflection Removal. arXiv preprint arXiv:1805.11589, 2018.
- Qian R., Tan R. T., Yang W., Su J. and Liu J. Attentive Generative Adversarial Network for Raindrop Removal from A Single Image. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018.