Privately Owned Vehicle Work Group Meeting - 2024/12/23 - Slot 2

[m-zain-khawaja]

Agenda

• Update on SuperDepth Training
• Update on PathDet dataset curation
• Update on LaneDet dataset curation
• SceneSeg Data Collection results by AutoCore.ai

Discussion

Update on SuperDepth training

@m-zain-khawaja - The dataset curation for SuperDepth has been completed. At the moment, work is in progress for code review of the parsing scripts, updating of the documentation of SuperDepth and refactoring of code/clean-up. The aim for this week is to complete the main training loop scripts, data augmentation, and to research the cost function which will be used to train the SuperDepth network.

PathDet dataset curation

We are on track to complete dataset parsing for the PathDet dataset with target completion of Jan 2025

• TuSimple - 6,394 training data samples
• CULane - 15K training data samples (exact number to be updated once data has been uploaded to Kaggle)
• CurveLane - (to be started)
• BDD100K - (parsing in progress by @sarun-hub)
• ROADWorks - (parsing in progress by @docjag)
• Comma2k19 - (parsing in progress by @siddas27)

Comments

@TranHuuNhatHuy has submitted a pull request to merge the parsing code for CULane, @m-zain-khawaja will review and aim to merge this week. @TranHuuNhatHuy will also upload the processed CULane dataset to Kaggle

@siddas27 has downloaded the Comma2k19 dataset and has successfully visualized the drivable ego mask in the camera image

@sarun-hub did a great effort to parse lanes from BDD100K to determine the drivable path, however, @m-zain-khawaja suggested to instead use the ego drivable-space annotation to derive the left/right ego-lanes rather than the explicit lane-level annotations in the dataset.

@devang-marvania is working on getting familiar with the TuSimple dataset to begin parsing for LaneDet

@docjag is currently working on parsing the ROADWork dataset

Overall - great progress from the team across all data parsing efforts!

LaneDet dataset curation

We are on track to complete dataset parsing for the LaneDet dataset with target completion of Feb 2025

• TuSimple - (parsing in progress by @devang-marvania)
• CULane - (to be started)
• CurveLane - (to be started)

SceneSeg bug-fix

The SceneSeg inference scripts did not work on computers that did not have CUDA/Nvidia GPU - this was fixed by Alex and his Pull request has been merged with the main branch:

Pull request

Attendees

• @m-zain-khawaja (Senior Tech Lead at AWF and Privately Owned Vehicle Work Group Chair)
• Mohammad Musa (Chair of Safety Assurance WG at AWF and CEO of DeepenAI)
• David Walmroth (Co-Chair of OpenADKit Work Group at AWF and US Director of Business and Project Development at PIX Moving)
• @TranHuuNhatHuy (Masters Robotics Student at Tokyo Institute of Technology)
• @sarun-hub (Undergraduate student at Nagoya University)
• @docjag (Senior Computer Vision and Machine Learning Engineer)
• @siddas27 (ADAS Engineer at Canoo)
• @devang-marvania (Lead Engineer - User Experience, Calibration at Gentherm)
• Aadarsh Tiwari (Software Engineer at KPIT)
• Pranit Bhatt (Robotics Software Developer)
• Adeesh Desai (M.S.E Robotics Student at University of Purdue)
• Fernando Preto (PhD Researcher at University of São Paulo focused on ADAS)
• Mahmoud Dahmani (Robotics Software Engineer at Vecna Robotics)

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.