【数据科学赛】Democratising aerial navigation via robust and data-scalable computer vision. #CV #$10,000_french national institute of geographical and fore-优快云博客

CompHub 实时聚合多平台的数据类(Kaggle、天池…)和OJ类(Leetcode、牛客…）比赛。本账号会推送最新的比赛消息，欢迎关注！

更多比赛信息见 CompHub主页[1] 或点击文末阅读原文

以下内容摘录自比赛主页

Part1赛题介绍

题目

No GPS no problem! Democratising aerial navigation via robust and data-scalable computer vision.

主办方

French National Institute of Geographical and Forest Information (IGN).

举办平台

CodaLab

背景

We believe that the real-challenge of today's CV is to develop and validate a design philosophy for algorithms that can work robustly and in an economically/data scalable way. In other words:

Being able to train with limited or even no real/domain specific data (zero-shot). Those leveraging synthetic data generated by open source tools such as TOPO-DataGen.
Capable of producing accurate output with trustworthy uncertainty statements that can capture accurately the prediction errors.

Part2时间安排

Start: Nov. 21, 2022, 10 a.m.
End: March 21, 2023, 11:59 p.m.

Part3奖励机制

Do you believe like us in pushing CV advances in the real world? Will you be a champion in proving CV advances benefits to the world of aviation? Then join us and earn:

The prize: 10 000 CHF (8 000 CHF for 1st place and 2 000 CHF for second) based on the leaderboard ranking. Winners commit to share their solution (git and model explanation report) open source for the benefit of all.
Respect: Being the leader in the real-world challenge of aerial navigation! Leaderboard open for all with no restrictions.

Part4赛题描述

Today aerial autonomous systems for navigation and control are heavily dependent on the robustness of GNSS reception. Lack of thereof (terrain, weather, or adversarial spoofing) can lead to loss of autonomous system absolute orientation. This in the medium to long run (error will always accumulate in time with dead reckoning) can be unsafe for operation, especially on beyond line of sight missions. Despite significant progress in Computer Vision, most learning-based approaches target a single domain and require a dense database of geo-tagged images to function well. Or at least a calibration set of real images taken closely from the domain of operation. Several industry attempts are made in the same direction, however, understandably without an official benchmark or validation.

We challenge you to prove your approach can work robustly and in an economically/data scalable way. To do this by proving you can compute accurate 6D camera poses with known uncertainty on our challenge validation dataset. To do this by having only access to: