文章简介
论文链接 代码链接
Accepted at ECCV2020
Main contribution: mutli-modal distillation unit & feature propagation module; 可以结合多个任务在不同scale下的特点结果.
任务:(室内)场景理解 scene parsing; 辅助任务: depth, semantic segmentation
Background & motivation
- 多任务相比于Single task的优越性:相对于single task,更节省memory;因为减少了重复的feature计算所以有increased speed;如果有complementary information会效果更好
-
First, due to their layer sharing, the resulting memory footprint is substantially reduced. Second, as they explicitly avoid to repeatedly calculate the features in
the shared layers, once for every task, they show increased inference speeds. Most importantly, they have the potential for improved performance if the associated tasks share complementar