ETHZ DLAD 2021 Exercise 2: Multitask Learning

Secret url: https://competitions.codalab.org/competitions/30245?secret_key=dac097ab-d5e9-42a3-8c12-bdcf4acd1f4f
Organized by anton - Current server time: April 4, 2025, 7:56 p.m. UTC

Current

Testing
March 22, 2021, midnight UTC

End

Competition Ends
Never

Welcome to the DLAD 2021 exercise 2: Multitask Learning competition! Please refer to the course PDF for instructions how to get data, train models, prepare submissions, and submit them to the leaderboard.

To make a submission, first train your model as explained in the assignment PDF, and make sure that a 'submission.zip' archive was generated in the experiment log directory. Next, navigate to the PARTICIPATE tab, select SUBMIT RESULTS, then enter an meaningful description of the submission, click SUBMIT button, and select the 'submission.zip' file from the log directory of your experiment. Multiple submissions can be uploaded, each will be graded, but only one will be shown in the leaderboard (you can choose which one). All three scores should be reported for each meaningful change of the model, which caused results improvement.

The Miniscapes dataset (this dataset) is a derivative work of the Synscapes dataset. Synscapes license agreement can be found at https://7dlabs.com/synscapes-license. This dataset is meant for educational purposes only, and must not be displayed, redistributed, modified, used for commercial purposes, stored, or retained after the end of this competition, assignment, or the course, whichever happens first. By enrolling into this challenge or downloading the training and validation splits, the user agrees with these terms and conditions, as well as with the license agreement of the original dataset. The user agrees additionally to take all necessary precautions to prevent any violation of all related conditions, and erase all copies of all parts of this dataset from all private, public, and temporary machines, cold storage, shared storage, where the aforementioned parts were placed by the user or his programs.

Testing

Start: March 22, 2021, midnight

Description: IoU (intersection-over-union) is a metric of performance of the semantic segmentation task, its values lie in the range [0,100], higher values are better. SI-logRMSE (scale-invariant log root mean squared error) is a metric of performance of the monocular depth prediction task, its values are positive, lower values are better. The Multitask metric is a simple product of the task-specific metrics, computed as max(iou-50, 0) + max(50-silogrmse, 0) , its values lie in the range [0,100], higher values are better. The leaderboard can be sorted by each of the metrics by clicking the column header, but only Multitask metric is used for ranking submissions.

Competition Ends

Never

You must be logged in to participate in competitions.

Sign In
# Username Score
1 lechen 77.16
2 kashen 74.46
3 yunkao 74.46