PASCAL in Detail

Make PASCAL great again!

CVPR 2017 Workshop Challenge

Honolulu, Hawaii, USA

PASCAL in Detail Workshop Challenge, July 26th (full day) 2017 @ Honolulu, Hawaii, USA

The last decade has seen enormous improvement for tasks such as object detection due to the creation of large annotated datasets which enable learning and benchmarking object detection algorithms. The goal of this workshop challenge is to measure the progress in image understanding as reflected in a diverse set of visual tasks, capturing both low- and high-level aspects of the vision problem. We thereby aspire to promote research that pushes the performance envelope in all facets of current computer vision research. To accomplish this we intend to systematize the evaluation of a large selection of representative visual tasks. This challenge provides a dataset with multiple-track detailed per-pixel labelings on PASCAL VOC images. For all datasets we will present baseline results for researchers to compare to. We expect that this will act like a catalyst for promoting research in multi-task learning, which we consider to be a key direction for future vision research.

Keynote speakers

Alan L. Yuille (Bloomberg Distinguished Professor at Johns Hopkins Univ.)

Larry Zitnick (Research Manager at Facebook AI Research)​

Kaiming He​ ​ (Research Scientist at Facebook AI Research)

Larry Zitnick
Kaiming He​

Tentative Schedule

9:00 - 9:30 Introduction and welcome (A. Yuille)

9:30 - 10:30 Single-track challenges and benchmarks (All)

10:30 - 12:00 Winning entries presentations

12:00 - 13:00 Invited keynote I (A. Zisserman)

13:00 - 14:00 Lunch Break

14:00 - 14:40 Invited keynote II (L. Zitnick)

14:00 - 14:30 Multi-track challenges (A.Vedaldi, I.Kokkinos)

14:30 - 15:30 Winning entries presentations

15:30 - 16:00 Coffee break

16:00 - 17:00 Invited keynote III (K. He)

17:00 - 17:30 Closing remarks and panel discussion

Tracks

  1. Object Detection
  2. Instance Segmentation
  3. Semantic Segmentation
  4. Part Segmentation
  5. Boundary Detection
  6. Occlusion Recognition
  7. Saliency Detection
  8. Line Segment Detection

Notably, apart from letting entries compete on the individual tracks, we will promote entries that aim at simultaneously attacking multiple tracks, using multi-task approaches. Entries that perform multiple tasks will be “gathering medals” across them - and the ranking among methods will be based on the sum “gold”, “silver” and “bronze” medals that they gather, in the same way that nations are ranked in the Olympics.

Organizers and Technical Staffs

Sanja Fidler
Iasonas Kokkinos
Roozbeh Mottaghi
George Papandreou
Raquel Urtasun
Andrea Vedaldi
Alan L. Yuille

In alphabetical order of last name

Please do not hesitate to contact iasonas.kokkinos@gmail.com regarding any issue about this challenge.