maskformer_panoptic_R50_bs64_554k.yaml

_BASE_: Base-COCO-PanopticSegmentation.yaml
MODEL:
  META_ARCHITECTURE: "MaskFormer"
  SEM_SEG_HEAD:
    NAME: "MaskFormerHead"
    IN_FEATURES: ["res2", "res3", "res4", "res5"]
    IGNORE_VALUE: 255
    NUM_CLASSES: 133
    COMMON_STRIDE: 4  # not used, hard-coded
    LOSS_WEIGHT: 1.0
    CONVS_DIM: 256
    MASK_DIM: 256
    NORM: "GN"
    # add additional 6 encoder layers
    PIXEL_DECODER_NAME: "TransformerEncoderPixelDecoder"
    TRANSFORMER_ENC_LAYERS: 6
  MASK_FORMER:
    TRANSFORMER_IN_FEATURE: "transformer_encoder"
    DEEP_SUPERVISION: True
    NO_OBJECT_WEIGHT: 0.1
    DICE_WEIGHT: 1.0
    MASK_WEIGHT: 20.0
    HIDDEN_DIM: 256
    NUM_OBJECT_QUERIES: 100
    NHEADS: 8
    DROPOUT: 0.1
    DIM_FEEDFORWARD: 2048
    ENC_LAYERS: 0
    DEC_LAYERS: 6
    PRE_NORM: False
    # COCO model should not pad image
    SIZE_DIVISIBILITY: 0
    TEST:
      PANOPTIC_ON: True
      OVERLAP_THRESHOLD: 0.8
      OBJECT_MASK_THRESHOLD: 0.8