detectron相关

一.主要流程

1.训练命令

python tools/train_net.py \
    --cfg detectron/datasets/data/experiment/config/e2e_faster_rcnn_R-50-FPN_1x.yaml \
    OUTPUT_DIR detectron/datasets/data/experiment/result

• 复制进去的时候用sublime或者其他正式的编辑器编辑，否则空格容易出错。
• /Detectron/detectron与detectron是相同作用的，前者是绝对路径，后者是相对路径。

1.1 yaml文件中的相关说明

MODEL:
  TYPE: generalized_rcnn
  CONV_BODY: FPN.add_fpn_ResNet50_conv5_body
  NUM_CLASSES: 5
  FASTER_RCNN: True
NUM_GPUS: 1
SOLVER:
  WEIGHT_DECAY: 0.0001
  LR_POLICY: steps_with_decay
  BASE_LR: 0.02
  GAMMA: 0.1
  #MAX_ITER: 90000
  #STEPS: [0, 60000, 80000]
  MAX_ITER: 1000
  STEPS: [0, 600, 800]

FPN:
  FPN_ON: True
  MULTILEVEL_ROIS: True
  MULTILEVEL_RPN: True
FAST_RCNN:
  ROI_BOX_HEAD: fast_rcnn_heads.add_roi_2mlp_head
  ROI_XFORM_METHOD: RoIAlign
  ROI_XFORM_RESOLUTION: 7
  ROI_XFORM_SAMPLING_RATIO: 2
TRAIN:
  WEIGHTS: https://dl.fbaipublicfiles.com/detectron/ImageNetPretrained/MSRA/R-50.pkl
  DATASETS: ('coco_2014_train', 'coco_2014_valminusminival')
  SCALES: (800,)
  MAX_SIZE: 1333
  BATCH_SIZE_PER_IM: 512
  RPN_PRE_NMS_TOP_N: 2000  # Per FPN level
  SNAPSHOT_ITERS: 200
TEST:
  DATASETS: ('coco_2014_minival',)
  SCALE: 800
  MAX_SIZE: 1333
  NMS: 0.5
  RPN_PRE_NMS_TOP_N: 1000  # Per FPN level
  RPN_POST_NMS_TOP_N: 1000
OUTPUT_DIR: .

• MAX_ITER:最大迭代次数
• SNAPSHOT_ITERS:间隔多少次保存一次权重

2.测试命令

python tools/infer_simple.py \
    --cfg detectron/datasets/data/experiment/config/e2e_faster_rcnn_R-50-FPN_1x.yaml \
    --output-dir detectron/datasets/data/experiment/result\
    --image-ext jpg \
    --wts detectron/datasets/data/experiment/result/train/coco_2014_train:coco_2014_valminusminival/generalized_rcnn/model_final.pkl \
    detectron/datasets/data/coco/coco_val2014

• image-ext：测试时，读取的文件的格式
• wts：训练好的权重的位置，定位到pkl文件
• 最后一行是demo，就是我们需要验证的数据