2nd Workshop on Traditional Computer Vision in the Age of Deep Learning (TradiCV)

Slides of the talks

• Fredrik Kahl (pdf)

• Laurent Kneip (pdf)

• Vincent Leroy (pdf)

• Richard Szeliski (pdf)

In the last 5-10 years we have witnessed that deep learning has revolutionized Computer Vision,  conquering the main scene in most vision conferences. However, a number of problems and topics for which deep-learned solutions are currently not preferable over classical ones exist, that typically involve a strong mathematical model (e.g., camera calibration and structure-from-motion).

This workshop concentrates on algorithms and methodologies that address Computer Vision problems in a “traditional” or “classic” way, in the sense that analytical/explicit models are deployed, as opposed to learned/neural ones. A particular focus will be given to traditional approaches that perform better than neural ones (for instance, in terms of generalization across different domains) or that, although performing sub-par, provide clear advantages with respect to deep learning solutions (for instance, in terms of efforts to collect data, computational requirements, power consumption or model compactness).  

This workshop also encourages critical discussions about preferring a traditional solution rather than a deep learning approach and also explores relevant questions about how to bridge the gap between learning and classic knowledge. We also expect an insightful discussion about ethical implications of traditional vision in comparison to deep learning approaches.

• 2.00pm - Introduction and welcome by the organizers

• 2.10pm - Spotlight Session 

• 2.50pm - Invited talk #1 - Fredrik Kahl 

• 3.30pm - Coffee break and poster session

• 4.00pm - Invited talk #2 - Laurent Kneip

• 4.40pm - Invited talk #3 - Vincent Leroy

• 5.20pm - Invited talk #4 - Richard Szeliski

• 6.00pm - Closing