Feature Pyramid Network for Object Detection

Paper Link : https://arxiv.org/pdf/1612.03144.pdf

Note : Our code has been merged into official gluon-cv, we recommend the official version, details can be found: dmlc/gluon-cv#494

This repo is an unofficial implementation of FPN in a gluon-cv style, we will follow up with the official code and please refer to the official tutorials for more information.

Usage

1.Install the mxnet

• pip install mxnet-cu92 --pre --upgrade

2.Clone this code and run the script *

Results

VOC07

• e2e_fastercnn_fpn_resnet50_v1b_1x_voc0712 : mAP=0.8035 on VOC07 Test.
  • train-log : https://drive.google.com/open?id=1obMRJedBHXrNPZLsahiR6WMQyYJFnli0
  • models : https://pan.baidu.com/s/1JrQYMLCPXuWGViSokS1LLg
• e2e_fastercnn_fpn_resnet101_v1d_1x_voc0712 : mAP=0.8301 on VOC07 Test.
  • train-log : https://drive.google.com/open?id=1Z7FeZ7fEuhYi1d4ExBlaLcJLl8RBSfFG

VOC12

• e2e_fastercnn_fpn_resnet50_v1d_1x_voc07++12 : Stay tuned!
  • train-log : https://drive.google.com/open?id=18NFrME-TM_8cvYTkKwtrqvcPCmZ5L-MG

COCO17

• e2e_fastercnn_fpn_resnet50_v1d_2x_coco17 : Stay tuned!

Acknowledgements

Below repos give me a lot of inspiration

• https://github.com/msracver/Deformable-ConvNets
• https://github.com/guoruoqian/FPN_Pytorch
• https://github.com/facebookresearch/Detectron