Reproduce “Semantic-Drive: Democratizing Long-Tail Data Curation via Open-Vocabulary Grounding and Neuro-Symbolic VLM Consensus” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Antonio Guillen-Perez
Semantic-Drive uses an open-vocabulary detector to guide vision-language models toward driving hazards, then checks scene descriptions with a consensus judge and rule-based verifier before indexing them. On 108 curated challenge frames, it reached 0.966 micro-averaged recall and 0.676 risk-score MAE, versus CLIP’s 0.475 recall and a single scout’s 1.130 MAE; results do not establish continuous-video performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.