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).

In total, there are three versions of CFENet:

1. CFENet (the master branch on the going)

2. CFENet+ (the working branch)

3. CFENet++ (under writing)

Citing

If you find this code useful in your research, please consider citing:

@misc{zhao2018cfenet,
  author = {Qijie Zhao},
  title = {CFENet: Object Detection with Comprehensive Feature Enhancement Module},
  howpublished = {\url{https://github.com/qijiezhao/CFENet}},
  year = {2018}