PUBLISHED ON OCT 2, 2016

Important information

Schedule

All times are PT (Pacific Time).

  • 8:30-8:45: Welcome
  • 8:45-9:30: Invited talk by Prof. Krystian Mikolajczyk (Imperial College London)
  • 9:30-10:15: Invited talk by Prof. Dieter Fox (University of Washington)
  • 10:15-10:30: Workshop paper, “Two-stage Discriminative Re-ranking for Large-scale Landmark Retrieval”, by Shuhei Yokoo (University of Tsukuba)
  • 10:30-10:45: Workshop paper, “Match or No Match: Keypoint Filtering based on Matching Probability”, by Alexandra Papadaki (Technical University of Berlin)
  • 10:45-11:15: Challenge introduction
  • 11:15-11:30: Challenge results and discussion
  • 11:30-13:45: Talks by challenge winners/participants (final schedule to be announced next week)
  • 13:45-14:00: Closing

About

Traditional, keypoint-based formulations for image matching are still very competitive on tasks such as Structure from Motion (SfM), despite recent efforts on tackling pose estimation with dense networks. In practice, many state-of-the-art pipelines still rely on methods that stood the test of time, such as SIFT or RANSAC.

In this workshop, we aim to encourage novel strategies for image matching that deviate from and advance traditional formulations, with a focus on large-scale, wide-baseline matching for 3D reconstruction or pose estimation. This can be achieved by applying new technologies to sparse feature matching, or doing away with keypoints and descriptors entirely.

The workshop topics include (but are not limited to):

  • Reformulating keypoint extraction and matching pipelines with deep networks.
  • Applying geometric constraints into the training of (sparse or dense) deep networks.
  • Leveraging additional cues such as semantics.
  • Developing adversarial methods to deal with conditions where current methods fail (weather changes, day versus night, etc.).
  • Exploring attention mechanisms to match salient image regions.
  • Integrating differentiable components into 3D reconstruction frameworks.
  • Matching across different data modalities such as aerial versus ground.

Organisers

Vincent
Lepetit
University of Bordeaux
Jiri
Matas
Czech Technical University
Dmytro
Mishkin
Czech Technical University
Johannes
Schönberger
Microsoft
Eduard
Trulls
Google
Kwang Moo
Yi
University of Victoria


The workshop will be held at CVPR 2020 in Seattle, WA, on June 19, 2020 (morning).

Talks

News

  • May 7: CVPR 2020 has moved to a fully virtual model. Stay tuned for details.
  • February 17: Confirming Krystian Mikolajczyk and Dieter Fox as speakers.
  • February 13: Added Call for Papers.
  • February 10: Updated website for the 2020 challenge.
  • January 24: Updated website for 2020 CVPR edition.

Important dates

  • Challenge publication: February 12, 2020
  • Paper submission deadline: March 31, 2020
  • Notification to authors: April 10, 2020
  • Camera-ready deadline: April 15, 2020 (papers must be archived by April 20)
  • Challenge submission deadline: see challenge website
  • Deadline for short descriptions on challenge submissions: see challenge website
  • Deadline for short abstract submission (non-archived posters): May 15, 2020
  • Workshop date: June 19, 2020 (half day: morning)

(All dates are at 11:59PM, Pacific Time.)

Challenge

We have renewed the challenge for last year, with many changes. The framework has been open-sourced, and we have greatly improved the evaluation server. For details, please refer to the challenge website.

Paper Submission

We welcome full paper submissions and extended abstracts, which will be presented in the workshop as posters. Please refer to the Call for Papers.

Contact

Please reach us with any questions at imagematching@uvic.ca.