MODEL:
  BACKBONE:
    FREEZE_AT: 0
    NAME: "build_resnet_backbone"
  WEIGHTS: "detectron2://ImageNetPretrained/torchvision/R-50.pkl"
  PIXEL_MEAN: [123.675, 116.280, 103.530]
  PIXEL_STD: [58.395, 57.120, 57.375]
  RESNETS:
    DEPTH: 50
    STEM_TYPE: "basic"  # not used
    STEM_OUT_CHANNELS: 64
    STRIDE_IN_1X1: False
    OUT_FEATURES: ["res2", "res3", "res4", "res5"]
    # NORM: "SyncBN"
    RES5_MULTI_GRID: [1, 1, 1]  # not used

SOLVER:
  IMS_PER_BATCH: 8
  BASE_LR: 0.0001
  STEPS: (260231, 283888)
  MAX_ITER: 295717
  WARMUP_FACTOR: 1.0
  WARMUP_ITERS: 10
  CHECKPOINT_PERIOD: 10000
  WEIGHT_DECAY: 0.05
  OPTIMIZER: "ADAMW"
  BACKBONE_MULTIPLIER: 0.1
  CLIP_GRADIENTS:
    ENABLED: True
    CLIP_TYPE: "full_model"
    CLIP_VALUE: 1.0
    NORM_TYPE: 2.0
  AMP:
    ENABLED: True
INPUT:
  IMAGE_SIZE: 768
  MIN_SCALE: 0.1
  MAX_SCALE: 2.0
  FORMAT: "RGB"
  MIN_SIZE_TRAIN: (1024,)
  MAX_SIZE_TRAIN: 1024
  DATASET_MAPPER_NAME: "coco_combine_lsj"
  MASK_FORMAT: "bitmask"
  COLOR_AUG_SSD: True

DATASETS:
  TRAIN: ("openvocab_coco_2017_train_panoptic_with_sem_seg",)
  TEST: ("openvocab_ade20k_panoptic_val",)  # to evaluate instance and semantic performance as well
DATALOADER:
  SAMPLER_TRAIN: "MultiDatasetSampler"
  USE_DIFF_BS_SIZE: False
  DATASET_RATIO: [1.0]
  DATASET_BS: [2]
  USE_RFS: [False]
  NUM_WORKERS: 8
  DATASET_ANN: ['mask']
  ASPECT_RATIO_GROUPING: True
TEST:
  EVAL_PERIOD: 10000
VERSION: 2