Skip to content
Hot eventLive

RBMatch:半监督建筑足迹提取的双层级类别重平衡

1 reports1 sources3 hr ago updated

Get the story

AI overview

2026年10月7日,arXiv Computer Vision(一手)报道了 RBMatch 框架。该工作面向半监督建筑足迹提取任务,提出双层级类别重平衡方法,联合调控伪标签生成与无监督优化两个阶段。研究指出,仅在伪标签选择阶段进行平衡时,背景偏差仍会在无监督损失优化过程中重现,作者将这一问题称为“imbalance leak”。RBMatch 通过在两个层级同时施加类别重平衡来应对该问题。报道未提供实验数据、基准结果或代码链接等进一步细节。

Generated from reports · updated 3 hr ago

Timeline

Follow the coverage from different angles.

Oct 7, 2026
  1. arXiv · Computer Vision
    RBMatch:面向半监督建筑足迹提取的双层级类别重平衡

    提出 RBMatch,一个面向半监督建筑足迹提取的双层级类别重平衡框架,联合调控伪标签生成与无监督优化,以解决仅平衡伪标签选择时背景偏差在无监督损失优化中重现的“imbalance leak”问题。

Heat trend

Current heat 9·Comparable peak 10(Oct 7)·Comparable change over 24 hours –

02.557.510Oct7Oct7Oct7Oct7

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.