Skip to content

Latest commit

 

History

History
112 lines (76 loc) · 4.56 KB

README.md

File metadata and controls

112 lines (76 loc) · 4.56 KB
Raw

Training COCO 2017 Object Detection and Segmentation via Learning Feature Pyramids

The code is provided by Guangrun Wang.

Sun Yat-sen University (SYSU)

Table of Contents

  1. Introduction
  2. Usage
  3. Citation

Introduction

This repository contains the training & testing code on COCO 2017 object detection and instance segmentation via learning feature pyramids (LFP). LFP is originally used for human pose machine, described in the paper "Learning Feature Pyramids for Human Pose Estimation" (https://arxiv.org/abs/1708.01101). We extend it to the object detection and instance segmentation.

Results

These models are trained on COCO 2017 training set and evaluated on COCO 2017 validation set. MaskRCNN results contain both bbox and segm mAP.

  • COCO Object Detection
Method MASKRCNN_BATCH resolution schedule AP bbox AP bbox 50 AP bbox 75
ResNet50 512 (800, 1333) 360k 37.7 57.9 40.9
Ours 512 (800, 1333) 360k 39.8 60.2 43.4
  • COCO Instance Segmentation
Method MASKRCNN_BATCH resolution schedule AP mask AP mask 50 AP mask 75
ResNet50 512 (800, 1333) 360k 32.8 54.3 34.7
Ours 512 (800, 1333) 360k 34.6 56.7 36.8

The schemes have the same configuration and mAP as the R50-C4-2x entries in Detectron Model Zoo.

Usage

  • The model is first pretrained on the ImageNet-1K, where the training scripts can be found Guangrun Wang's github. We also provide the trained ImageNet models as follows.

    Baidu Pan, code: zvgd

    Google Drive

  • Training script for COCO object detection and instance segmentation:

python3 train.py --load /home/grwang/seg/train_log_resnet50/imagenet-resnet-d50/model-510000    --gpu 0,1,2,3,4,5,6,7   --logdir mask-pyramid-train
  • Testing script for COCO object detection and instance segmentation:
python3 train.py  --evaluate output.json  --load /home/grwang/seg/train_log_resnet50/imagenet-resnet-d50/model-510000    --gpu 0,1,2,3,4,5,6,7   --logdir mask-pyramid-test

  • Trained Models of COCO:

    Model trained for evaluation on COCO 2017 object detection and instance segmentation task:

    Baidu Pan, code: w7o9

    Google Drive

Citation

If you use these models in your research, please cite:

@inproceedings{yang2017learning,
        title={Learning feature pyramids for human pose estimation},
        author={Yang, Wei and Li, Shuang and Ouyang, Wanli and Li, Hongsheng and Wang, Xiaogang},
        booktitle={The IEEE International Conference on Computer Vision (ICCV)},
        volume={2},
        year={2017}
    }

Dependencies

  • Python 3; TensorFlow >= 1.4.0 (>=1.6.0 recommended due to a TF bug);
  • pycocotools, OpenCV.
  • Pre-trained ImageNet model: google drive; baidu pan(code: zvgd).
  • COCO data. It needs to have the following directory structure:
DIR/
  annotations/
    instances_train2014.json
    instances_val2014.json
    instances_minival2014.json
    instances_valminusminival2014.json
  train2014/
    COCO_train2014_*.jpg
  val2014/
    COCO_val2014_*.jpg

minival and valminusminival can be download from here.

  • Tensorpack The code depends on Yuxin Wu's Tensorpack. For convenience, we provide a stable version 'tensorpack-installed' in this repository. 
    # install tensorpack locally:
cd tensorpack-installed
python setup.py install --user