PyTorch implementation of Single-Shot Refinement Neural Network for Object Detection. The official and original Caffe code can be found here.

Table of Contents

[TOC]

Performance

Train : VOC 07+12 trainval

Test: VOC 07 test

Methond	My mAP
base line	81.95%
mixed precision	81.85%
withmixup data augmentation	83.09%

Experiment Record

2019-5-21 18:47:58

1. Mixed precision training with apex
2. Mixup data augmentation with β(1.5, 1.5)

2019-5-18 18:08:16

1. add csv format . **csv format ** is an easy way to define your own dataset，not like the VOC format which is too complicated .

   train,val test with csv format like the following

   im_path, xmin, ymin, xmax, ymax, class_name

   eg:

   /root/data/07+12_trainval/000005.jpg,263,211,324,339,chair
   /root/data/07+12_trainval/000005.jpg,165,264,253,372,chair
   /root/data/07+12_trainval/000005.jpg,241,194,295,299,chair
   /root/data/07+12_trainval/007212.jpg,143,9,445,331,dog
   /root/data/07+12_trainval/002657.jpg,212,198,288,264,horse
   /root/data/07+12_trainval/002657.jpg,268,353,296,375,person
   

   Also , need a .csv file named classes.csv