fcos_R_50_FPN_1x_coco_pose2_full.yaml

MODEL:
  META_ARCHITECTURE: "GeneralizedRCNNPose"
  # WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50"
  WEIGHT: ""
  RPN_ONLY: True
  FCOS_ON: True
  POSENET_ON: True
  BACKBONE:
    CONV_BODY: "R-50-FPN-RETINANET"
  RESNETS:
    BACKBONE_OUT_CHANNELS: 256
  RETINANET:
    USE_C5: False # FCOS uses P5 instead of C5
  FCOS:
    NUM_CLASSES: 4  # COCO is 80+1, DangerDet is 3+1
  POSENET:
    N_STACK: 2  # the num of stacked hourglasses
    OUTPUT_DIM: 17  #  number of keypoints / heatmaps, COCO:17 MPII:16
    FUSE_TYPE: "LAST"  # default is "LAST", you can choose "FULL"
DATASETS:
  TRAIN: ("coco_dangerdet_Wpose_train",)  # all images shape imgh:imgw = 1600:2400 = 2:3
  # TEST: ("coco_dangerdet_Wpose_val",)
  TEST: ("coco_dangerdet_val",)
INPUT:
  # MIN_SIZE_TRAIN: (800,)  # input shape is 800:1200 = 2:3
  # MAX_SIZE_TRAIN: 1333
  # MIN_SIZE_TEST: 800
  # MAX_SIZE_TEST: 1333
  MIN_SIZE_TRAIN: (768,)  # 1/8 * 1/16, input shape is 768:1152 = 2:3, 768*1/8*1/16 = 6
  MAX_SIZE_TRAIN: 1280
  MIN_SIZE_TEST: 768
  MAX_SIZE_TEST: 1280
  # MIN_SIZE_TRAIN: (1024,)  # 1/8 * 1/16, input shape is 1024:1536 = 2:3, 1024*1/8*1/16 = 8
  # MAX_SIZE_TRAIN: 1664
  # MIN_SIZE_TEST: 1024
  # MAX_SIZE_TEST: 1664
DATALOADER:
  SIZE_DIVISIBILITY: 32
SOLVER:
  BASE_LR: 0.004  # 0.01
  WEIGHT_DECAY: 0.0001
  STEPS: (9000, 12000)
  MAX_ITER: 15000
  CHECKPOINT_PERIOD: 3500
  IMS_PER_BATCH: 12  # 16
  WARMUP_METHOD: "constant"
TEST:
  BBOX_AUG:
    ENABLED: False
    H_FLIP: True
    SCALES: (400, 500, 600, 700, 900, 1000, 1100, 1200)  # w/o AUG, mAP 0.605
    MAX_SIZE: 2000
    SCALE_H_FLIP: True
    # ENABLED: True
    # H_FLIP: True
    # SCALES: (512, 768, 1024, 1280)  # w/ AUG, mAP 0.533
    # MAX_SIZE: 2000
    # SCALE_H_FLIP: True