Privately Owned Vehicle Work Group Meeting - 2024/11/18 - Slot 2

[m-zain-khawaja]

Agenda

Welcome to new contributors

An outreach effort and call for participation was carried out to encourage new contributors to participate in the Privately Owned Vehicle work group, and new members will have the opportunity to briefly introduce themselves

Overview of latest technology developments

• Update on SuperDepth neural network as part of AutoSeg Foundation Model
• Update on literature review for Lane Detection

Attendees

• @m-zain-khawaja (Senior Tech Lead at AWF and Privately Owned Vehicle Work Group Chair)
• Christian John (Chair of Strategic Planning Committee at AWF and President at TierIV)
• @mitsudome-r (Chair of Technical Steering Committee at AWF)
• Samet Kütük (Chair of Marketing Steering Committee at AWF)
• M. Fatih Cirit (Software Architect at AWF)
• David Walmroth (Co-Chair of OpenADKit Work Group at AWF and US Director of Business and Project Development at PIX Moving)
• Tran Huu Nhat Huy (Masters Robotics Student at Tokyo Institute of Technology)
• Elvin Abdinli (Masters Informatics at TUM)

Discussion

• Introductions by attendees
• Run through of agenda items by @m-zain-khawaja
• Welcome to Elvin , a new joiner to the Privately Owned Vehicle Work Group
• A detailed presentation was shared @m-zain-khawaja on the latest work related to SuperDepth - a vision-based depth estimation system which is part of the AutoSeg Foundation Model, this included:
  • motivations behind creating SuperDepth as a robust all-obstacles detection system, serving as a core safety layer for autonomous vehicles in edge-case scenarios
  • overview of neural network architecture of SuperDepth and explanation of key components of the network architecture
  • overview of existing vision-based depth estimation methods including deep stereo-matching methods, zero-shot monocular methods, as well as their strengths and limitations
  • overview of proposed methodology for SuperDepth as a best of both worlds approach
  • overview of importance on robust prediction at depth boundaries between foreground/background scene elements for reliable downstream pointcloud processing
• review of datasets used for training the SuperDepth network
  • sharing of current progress in terms of neural network blocks that have been designed and coded-up, as well as progress in dataset curation, with the goal to complete dataset curation by next week and be able to begin training the network by mid-December, with the aim to have the fully trained network ready at the start of the New Year

Subsequently, Tran gave a detailed presentation regarding a literature review which he had been working on related on lane detection algorithms including:

• Best performing algorithms for various benchmarks
• Review of methodologies used in various methods
• Discussion of loss function related to Lane IoU for improved robustness of lane detection
• @m-zain-khawaja asked Tran to upload his presentation as a Google Slides shared file for commentary and dissemination, as well as recommendation for Tran to also benchmark FPS of various methods and classify detection type, e.g. segmentation-based, keypoint-based, or parametric curve-based
• Tran will aim to complete the literature review by next week's meeting.

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.