The two models use a shared backbone to extract features.
这两个模型使用共享主干来提取特征。
In multi-task learning, a shared backbone can learn general representations, while each task has its own head for specific predictions.
在多任务学习中,共享主干可以学习通用表征,而每个任务再配一个各自的“头部”来做特定预测。