解读flow-guided feature aggregation for video object detection

文章主要贡献点：

Flow-guided feature aggregation, an end-to-end framework for
video object detection.

Improve the per-frame features by aggregation of nearby features along the motion path, and thus improve the video recognition accuracy.

  Or improve the per-frame feature learning by temporal aggregation

数据库：ImageNet VID dataset

              3862 video snippet from the traning set

              555 snippets from the validation set

              Fully annotated

              30 object categories (a subset of the categories in the ImageNet DET dataset),

相关工作：

本文工作：

      1. the feature extraction network is applied on individual frames to produce the per-frame feature maps

       2. To enhance the features at a reference frame, an optical flow network  [flownet] estimates themotions between the nearby frames an the reference frame. 

       3. The feature maps from nearby frames are warped to the reference maps, as well as its own feature maps on the reference frame, areaggregated according to an        adaptive weighting network. 

        4. The resulting aggregated feature maps are then fed to the detection network to produce the detection result on the reference frame.

         System: Feature extraction + flow estimation + feature aggregation + detection