Dieses Dokuwiki verwendet ein von Anymorphic Webdesign erstelltes Thema.

Data challenges

2018-2019 (updated Nov. 19: schedule)

The challenge will be on pollutant prediction in Grenoble area, in collaboration with ATMO Auvergne-Rhône Alpes.

Milestones

See https://competitions.codalab.org/competitions/20430?secret_key=9c39a737-0965-4120-89a4-d8b6cc08eddb

  • Presentation and delivery of the training data set (Oct. 25, 2018)
  • Presentation of the starting kit and expected format of outputs (Nov. 30, 2018, IM2AG F321. Nov. 19, 8:45AM)
  • Development of some basic solution (baseline). Short oral presentation of the baseline solution and short description of it; 1-2 pages report (Dec. 13, 3:30PM). You may express some particular needs in terms of computational power for the intensive session.
  • Intensive session, partly supervised by tutors (Feb. 6-8). On Feb. 8 (AM): prediction on evaluation data set; (PM): defence of the whole project and final proposal with questions from teachers and other team's members. Report of 4 to 10 pages with experience feedback (description and comparison of the different approaches that were tried).
Schedule, teams and registration

See https://competitions.codalab.org/competitions/20430?secret_key=9c39a737-0965-4120-89a4-d8b6cc08eddb.
Please do not register before being advised to do so, since this version of the site is temporary and likely to be replaced by a newer one.

Grading rules
  • Evaluation: 1/3 score (ranking / performance), 1/3 report, 1/3 oral presentation.
Presentation and report

The presentation and report should contain a description of the different approaches you tried, even if some of them were discarded because of unsatisfactory results. The report has to be exhaustive on that point, while the presentation may focus on 2 or 3 of them. There should be associated values and comparisons of the metrics, and on execution times. Try to analyse the shortcomings of the methods, explain how you gradually overcame them, and what could be further improved in your final proposal.

The oral presentation should last from 18 to 20 minutes and will be followed by questions. You may not spend more than one slide / 1 min 30 on the problem description and data, since these are common to every team.

The report should be from 4 to 10 pages long. You have to deliver it just before the presentations.

Rankings

2017-2018 (updated Nov. 9: participants and team composition)

The challenge will be on audio-visual speaker diarization.

Registered students:
Teams
Milestones
  • (October 12) Presentation of the context by Radu Horaud at the Data science seminar.
    Slides of the introductory presentation by Radu Horaud
    Before the seminar, you have to read the two mentioned articles.
  • (around October 20) Delivering the training data set
  • (around November 9) One session to get familiar with the training data set, and some software for video exploration
  • (December 11, room Ensimag H101) Development of some basic solution (baseline). Short oral presentation of the baseline solution and short description of it (1-2 pages report). You may express some particular needs in terms of computational power for the intensive session.
  • (February 7-9, room IM2AG F108, from 9 AM to 5 PM) Intensive session, partly supervised by tutors. February 9 AM: prediction on evaluation data set. February 9 PM: defence of the whole project and final proposal with questions from teachers and other team's members. Report of 4 to 10 pages with experience feedback (description and comparison of the different approaches that were tried).
Data and basic scripts

The data can be downloaded from https://github.com/Stephlat/dataChallengePerception. This page also contains details about the data and basic scripts for data visualization.

Rules
Presentation and report

The presentation and report should contain a description of the different approaches you tried, even if some of them were discarded because of unsatisfactory results. The report has to be exhaustive on that point, while the presentation may focus on 2 or 3 of them. There should be associated values and comparisons of the metrics, and on execution times. Try to analyse the shortcomings of the methods, explain how you gradually overcame them, and what could be further improved in your final proposal.

The oral presentation should last from 18 to 20 minutes and will be followed by questions. Presentations will start at 1:30 PM (or maybe 2PM, depending on participant's constraints). You may not spend more than one slide / 1 min 30 on the problem description and data, since these are common to every team.

The report should be from 4 to 10 pages long. You have to deliver it just before the presentations.

Please have a close-to-final version of both reports and slides on Thursday evening.

Submission of your predictions

The link to the test data is available on https://github.com/Stephlat/dataChallengePerception
You have to submit your predictions before 11:00 AM by sending a .zip archive containing 4 folders named Scenario03-MeetingRoom04 Scenario05-04 Scenario01-03 and Scenario05-03. In each of these, you must place your detection.txt files (but no .avi files).
Join to your archive the exact python code (whole set of scripts) that has been used to produce the detection.txt files.

Send your email Jean-Baptiste.Durand@univ-grenoble-alpes.fr, guillaume.delorme@inria.fr and stephane.lathuiliere@inria.fr.

The final score used to ranked teams is the mean MOTSA on the 4 videos, weighted by video lengths.

Send your report and presentation (PDF format only) before 1:30 PM at Jean-Baptiste.Durand@univ-grenoble-alpes.fr, guillaume.delorme@inria.fr, stephane.lathuiliere@inria.fr, li.liu@gipsa-lab.grenoble-inp.fr, olivier.michel@gipsa-lab.grenoble-inp.fr, Massih-Reza.Amini@univ-grenoble-alpes.fr and Radu.Horaud@inria.fr.

Rankings
  1. [0.805] Oussama.Zerguine@grenoble-inp.org, sebastien.rimbaud@phelma.grenoble-inp.fr, Ivan.Iudintsev@grenoble-inp.org, gregoire.mugnier@phelma.grenoble-inp.fr, mariia.garkavenko@etu.univ-grenoble-alpes.fr
  2. [0.797] jaime-enrique.romero-zuleta@grenoble-inp.org, ivan.koval@etu.univ-grenoble-alpes.fr, anastasiia.doinychko@etu.univ-grenoble-alpes.fr, ismail.arab.eng@gmail.com
  3. [0.745] Marvin.Lerousseau@grenoble-inp.org, Manuel-David Camargo-Rivera Manuel-David.Camargo-Rivera@grenoble-inp.org, kuntimakiala michel kuntima-kiala.michel@phelma.grenoble-inp.fr, Quoc Trung Vuong quoc-trung.vuong@grenoble-inp.org
  4. [0.743] Yassine.Laguel@grenoble-inp.org, Roman.Bresson@grenoble-inp.org, selim.chraibi@grenoble-inp.org, duc-anh.luu@grenoble-inp.org, nejim.zeineb@gmail.com
  5. [0.709] Jacqueline Sime jsime1559@gmail.com, Nour Serhan nour-taki@hotmail.com, Artem.Betlei@etu.univ-grenoble-alpes.fr, Aleksandra.Malkova@etu.univ-grenoble-alpes.fr
  6. [no score available] moradi.siamak324@gmail.com, vincent.nebot@phelma.grenoble-inp.fr, Parivash.Mohammadi-Khajehdehi@etu.univ-grenoble-alpes.fr, Vincent.Xuereb@grenoble-inp.org

Detailed results

collab/data_challenge.txt · Last modified: 2018/12/13 09:40 by jbdurand
Dieses Dokuwiki verwendet ein von Anymorphic Webdesign erstelltes Thema.
www.chimeric.de Valid CSS Driven by DokuWiki do yourself a favour and use a real browser - get firefox!! Recent changes RSS feed Valid XHTML 1.0