Privately Owned Vehicle Work Group Meeting - 2024/12/09 - Slot 1

[m-zain-khawaja]

Agenda

• Update on SuperDepth Dataset
• Combining PathDet and DiversionDet networks in AutoSeg
• Data curation update related to new PathDet network

SuperDepth Dataset Update

• MUAD - 3912 training data samples
• UrbanSyn - 6,732 training data samples
• KITTI - 4,325 training data samples
• Driving Stereo (found to be not suitable)
• Argoverse - 1,324106 training data samples
• MUSES - 25002,500 training data samples
• DDAD (to be started, download complete)
• BOREAS (to be started, download complete)

@m-zain-khawaja shared a detailed update on the datasets which have been parsed and the methods used to parse and process the latest datasets, Argoverse and MUSES.

PathDet Implementation Discussion

It has been decided to being work on implementing the PathDet branch before the LaneDet branch to ensure a robust pre-training of the PathContext block and to ensure that it can learn generalisable features related to both drivable paths and lanes, rather than lanes alone. Due to a lack of training data for DiversionDet, the DiversionDet neural head will be merged with PathDet such that there are two path prediction systems:

• LaneDet: predicts the positions of lanes
• PathDet: predicts a single drivable path under all driving conditions even in the absence of lanes, similar to end-to-end approaches

@m-zain-khawaja discussed the motivation behind the update to the AutoSeg architecture as well and the mechanism by which PathDet and DiversionDet can be merged, and the datasets that can be used to train the new PathDet network, including new datasets such as ROADwork dataset and Comma2k19.

PathDet dataset curation

@TranHuuNhatHuy has begun work on dataset curation starting with TuSimple dataset and shared his latest code updates and presentation slides related to his work on dataset curation.

Attendees

@m-zain-khawaja (Senior Tech Lead at AWF and Privately Owned Vehicle Work Group Chair)
Simon Thompson (Director of Engineering, Research at Tier IV)
@mitsudome-r (Chair of Autoware Technical Steering Committee)
@TranHuuNhatHuy (Masters Robotics Student at Tokyo Institute of Technology)
Lucus Xingang Liu (Autocore.ai)
Alexander Korvyakov (Senior Product Manager - Autonomous Vehicle Foundation at MOIA, part of VW Group)
Siddharth Das (ADAS Engineer at Canoo)
Devang Marvania (Lead Engineer - User Experience, Calibration at Gentherm)
Prerana Kolipaka (MS ECE at Georgia Tech)

Zoom Meeting Video Recording

Video Meeting Link
Please contact the work group lead (@m-zain-khawaja ) to request access to a recording of this meeting.