Claude Code-Driving Scenario Mining for the Argoverse 2 Challenge
2026-06-08 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors describe their approach to a challenge involving scenario mining in autonomous driving data. Their method involves four main steps: first, they use an AI agent called Claude Code to generate code automatically; second, they carefully select training examples based on a performance threshold; third, they have another AI session review the code for quality; and finally, they use another AI model (Qwen3-VL) to verify the scenes and remove errors. They tested their system using the Argoverse 2 dataset and reported the results.
autonomous drivingscenario miningClaude Code agentGLM 5.1few-shot learningcode reviewQwen3-VLArgoverse 2 datasetbalanced accuracy
Authors
Wei Deng, Caoshengzhe Xue, Shuaikun Liu, Zhaohong Liu, Mengshi Qi, Huadong Ma
Abstract
We present our submission to the CVPR 2026 Argoverse 2 Scenario Mining Challenge. Our system uses a four-stage pipeline: (1) autonomous code generation via a Claude Code agent powered by GLM~5.1, (2) iterative training set screening with Timestamp Balanced Accuracy threshold 0.8 to curate few-shot examples, (3) semantic code review by a separate Claude Code session, and (4) Qwen3-VL scene-level verification to filter false positives. We report results on the Argoverse 2 test set.