Skip to content

Latest commit

 

History

History
40 lines (34 loc) · 1.52 KB

README.md

File metadata and controls

40 lines (34 loc) · 1.52 KB

SSD project based on caffe

Deps

Usage:

  • create soft link of some bin files(caffe, convert_annoset, get_image_size) with caffe-ssd project
  • export PYTHONPATH=$SSD_CAFFE_ROOT/python:$PYTHONPATH
  • write labelmap_voc.prototxt files
  • run with the continuously num files
  • 1_createXml.py: create xml formats from origin labels or use labelImg tools to get the xml format labels
  • 2_createTrainVal.py: generate the test and trainval image name lists in ImageSets/Main, generate trainval test file lists and get test image size
  • [3_create_list.sh]: optional, 2_createTrainVal.py, generate the test, trainval image lists and get test image size
  • 4_create_data.sh: get label map and generate LMDB datas
  • 5_ssd_run.py: run ssd training and get the solver.prototxt and train_net
  • 6_ssd_run_direct.sh: training the net directly based on the train_net have generated

Note:

  • testLabelImgs/JPEGImages: the jpeg images
  • testLabelImgs/labels: the labels for each image
  • testLabelImgs/Annotations: the labels with .xml format
  • testLabelImgs/ImageSets/Main: the test and trainval txt files
  • the original labels is created by BBox-Label-Tool, the format is:
    object_num
    className x1min y1min x1max y1max
    className x2min y2min x2max y2max

Output Dirs

  • models:
  • examples:
  • job:
  • results:
  • data/test.txt
  • data/trainval.txt
  • data/test_name_size.txt

Net Instruction