Researchers have developed CoRE, a novel framework for weakly supervised learning that extracts fine-grained risk evidence from coarse video-level predictions. This method trains a video-level predictor and then uses structured interventions to measure how candidate temporal regions or entities affect the prediction. The resulting targets are distilled into a student model that can directly predict temporal and entity support without needing costly fine-grained annotations. CoRE has demonstrated effectiveness across driving video risk assessment, traffic anomaly localization, and general anomaly detection benchmarks. AI
IMPACT This framework could reduce the need for expensive fine-grained annotations in video analysis tasks.
RANK_REASON The cluster contains a research paper detailing a new learning framework. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
