wangwang110 / caffe-yolo9000

Caffe for YOLOv2 & YOLO9000

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

CAFFE for YOLO9000

Reference

YOLO9000: Better, Faster, Stronger

http://pjreddie.com/yolo9000/

https://github.com/yeahkun/caffe-yolo

Usage

caffe

   vim Makefile.config
   make

Data preparation

  cd data/yolo
  mkdir lmdb
  ln -s /your/path/to/VOCdevkit/ .
  python ./get_list.py
  # change related path in script convert.sh
  ./convert.sh 

Train

  cd examples/yolo/darknet_v3
  # change related path in script train.sh
  mkdir models
  ./train_darknet_v3.sh

Test a image

   cd examples/yolo/eval_detection
   jupyter notebook
   test_det.ipynb

Eval VOCtest2007(The first way online)

  # mAP reach ~56. Because of I train net poorly...you can try.
  cd examples/yolo/darknet_v3
  ./test_darknet_v3.sh

MODEL=./models/gnet_yolo_region_darknet_v3_pretrain_rectify_iter_200000.caffemodel

model is here:

https://pan.baidu.com/s/1nvHggFB t7ui

Eval VOCtest2007(The second way offline)

  cd examples/eval_detection
  python test_yolo_v2.py

Draw loss figure(avg_obj, avg_noobj, avg_class, avg_iou, recall)

  cd tools/yolo_extra
  python parse_log_yolo.py ./log/train_darknet_anchor.log ./log

If you want to train your datasets. you should edit the train_prototxt.

   [conv_reg layer] num_output = num * (num_class + coords + 1) = 5 * (your_classes_num + 4 + 1)
   [det_loss layer] num_class = your_classes_num
yolo9000-Tree example prototxts, model and .sh will update soon!

About

Caffe for YOLOv2 & YOLO9000

License:Other


Languages

Language:C++ 80.1%Language:Python 8.0%Language:Cuda 5.5%Language:CMake 2.7%Language:Protocol Buffer 1.7%Language:MATLAB 0.9%Language:Makefile 0.7%Language:Shell 0.4%Language:M 0.0%