A Context and Level Aware Feature Pyramid Network for Object Detection with Attention Mechanism

By Hao Yang, Yi Zhang

This project is based on mmdetection

Introduction

An object detection task includes classification and localization, which require large receptive field and high-resolution input respectively. How to strike a balance between the two conflicting needs remains a difficult problem in this field. Fortunately, feature pyramid network (FPN) realizes the fusion of low-level and high-level features, which alleviates this dilemma to some extent. However, existing FPN based networks overlooked the importance of features of different levels during fusion process. Their simple fusion strategies can easily cause overwritten of important information, leading to serious aliasing effect. In this paper, we propose an improved object detector based on context and level aware feature pyramid networks. Experiments have been conducted on mainstream datasets to validate the effectiveness of our network, where it exhibits superior performances than other state-of-the-art works.

Install

Please refer to INSTALL.md for installation.

note: In this project, we only uploaded the core configuration file and model, other files could be found in mmdetection.

Prepare data

  mkdir -p data/coco
  ln -s /path_to_coco_dataset/annotations data/coco/annotations
  ln -s /path_to_coco_dataset/train2017 data/coco/train2017
  ln -s /path_to_coco_dataset/test2017 data/coco/test2017
  ln -s /path_to_coco_dataset/val2017 data/coco/val2017

Training

./tools/dist_train.sh <CONFIG_FILE> <GPU_NUM> --validate --work_dir <WORK_DIR>

For example,

./tools/dist_train.sh configs/faster_rcnn_r50_clfpn_1x_coco.py 8 --validate --work_dir faster_rcnn_r50_clfpn_1x

see more details at mmdetection

Testing

python tools/test.py <CONFIG_FILE> <CHECKPOINT_FILE> --gpus <GPU_NUM> --out <OUT_FILE> --eval <EVAL_TYPE>

When test results of detection, use --eval bbox. When test results of instance segmentation, use --eval bbox segm. See more details at mmdetection.

For example,

python tools/test.py configs/faster_rcnn_r50_clfpn_1x_coco.py <CHECKPOINT_FILE> --gpus 8 --out results.pkl --eval bbox segm

Results on MS COCO testdev2017

Backbone	detector	schedule	mAP(det)
ResNet-50	Faster R-CNN	1x	39.2
ResNet-101	Faster R-CNN	1x	41.0
ResNeXt-101-64x4d	Faster R-CNN	1x	43.3

License

This project is released under the Apache 2.0 license