在训练阶段,shared featuremap进了detection branch之后得到的scoremap和geomap没有输入到roirotate去算recognitionloss,而是单独计算detection loss,recognition loss 用groundtruth输入到roiratate来计算
FOTS: Fast Oriented Text Spotting with a Unified Network训练阶段
最新推荐文章于 2022-11-04 15:07:23 发布