queryrcnn.cascade.res50.300pro.3x.yaml

MODEL:
  META_ARCHITECTURE: "QueryRCNN"
  WEIGHTS: "ImageNetPretrained/torchvision/R-50.pkl"
  MASK_ON: False
  PIXEL_MEAN: [123.675, 116.280, 103.530]
  PIXEL_STD: [58.395, 57.120, 57.375]
  BACKBONE:
    NAME: "build_retinanet_resnet_fpn_backbone"
  RESNETS:
    OUT_FEATURES: ["res2", "res3", "res4", "res5"]
    DEPTH: 50
    STRIDE_IN_1X1: False
  FPN:
    IN_FEATURES: ["res2", "res3", "res4", "res5"]
  RPN:
    IN_FEATURES: ["p3", "p4", "p5", "p6", "p7"]
  ROI_HEADS:
    IN_FEATURES: ["p2", "p3", "p4", "p5"]
  ROI_BOX_HEAD:
    POOLER_TYPE: "ROIAlignV2"
    POOLER_RESOLUTION: 7
    POOLER_SAMPLING_RATIO: 2
  QueryRCNN:
    RCNNHead: rcnnhead_wmha
    WITH_POS: False
    RPN:
      RPN_TYPE: "anchor_free"
      FPN_STRIDES: [8, 16, 32, 64, 128]
      NUM_CLASSES: 1
  SparseRCNN:
    NUM_HEADS: 2
    BBOX_WEIGHTS: [2., 2., 1., 1.]
    NUM_PROPOSALS: 300
    NUM_CLASSES: 80
    GIOU_WEIGHT: 2.
    CLASS_WEIGHT: 2.
SOLVER:
  IMS_PER_BATCH: 16
  BASE_LR: 0.000025
  STEPS: (210000, 250000)
  MAX_ITER: 270000
  WARMUP_FACTOR: 0.01
  WARMUP_ITERS: 1000
  WEIGHT_DECAY: 0.0001
  OPTIMIZER: "ADAMW"
  BACKBONE_MULTIPLIER: 1.0  # keep same with BASE_LR.
  CLIP_GRADIENTS:
    ENABLED: True
    CLIP_TYPE: "full_model"
    CLIP_VALUE: 1.0
    NORM_TYPE: 2.0
SEED: 40244023
DATASETS:
  TRAIN: ("coco_2017_train",)
  TEST: ("coco_2017_val",)
INPUT:
  MIN_SIZE_TRAIN: (480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800)
  CROP:
    ENABLED: False
    TYPE: "absolute_range"
    SIZE: (384, 600)
  FORMAT: "RGB"
TEST:
  EVAL_PERIOD: 7330
DATALOADER:
  FILTER_EMPTY_ANNOTATIONS: False
  NUM_WORKERS: 4
VERSION: 2