/CFENet

Comprehensive Feature Enhancement Module for Single-Shot Object Detector

CFENet: Object Detection with Comprehensive Feature Enhancement Module

we will release code after the paper publicly available, which is accepted by ACCV2018 recently.

Single Model Records(Non-ensemble results):

  • 1st on UA-DETRAC

  • 6th on KITTI car detection

  • 2nd on WAD workshop

  • 1st on Visdrone object detection in videos

  • 6th on Wider Face for face detection

  • AP of 43.5 on COCO(ResNet101-CFENet512)

Note that CFENet is only an efficient one-stage detector, which can achieve 23+fps on MS-COCO when single-scale inference(VGG-CFENet800).

Now, we have opened the working branch, we wish you can try to train it with different configurations, this can help us find BUGs.

What's more, we are training the referenced models recently(CFENet300-SEResNet50, CFENet512-VGG16), Once it finished, we will open it at master branch.