{"id":"5a53491a-9a48-4313-aefe-2f7b4e317f32","arxiv_id":"2505.14014","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"EGFormer dynamically scores and drops the least useful sensor modality at each processing stage, cutting parameters by up to 91 percent and GFLOPs by half while keeping segmentation accuracy competitive.","lead":"EGFormer is a lightweight system for image understanding that combines several sensor types, such as cameras, depth and LiDAR, while dropping the least useful sensor at each step. It claims to match prior accuracy with far fewer parameters and to transfer from synthetic to real scenes without labels.","discovery_kind":"extension","skeptic_critique":null,"referee_report":null,"author_rebuttal":null,"desk_editor":null,"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-07T15:43:14.739840+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}