MGNet-Cityscapes-Fine.yaml

MODEL:
  META_ARCHITECTURE: "MGNet"
  WEIGHTS: "./weights/imagenet_weights.pkl"
  PIXEL_MEAN: [ 123.675, 116.280, 103.530 ]
  PIXEL_STD: [ 58.395, 57.120, 57.375 ]
  BACKBONE:
    FREEZE_AT: 0
    NAME: "build_resnet_iabn_backbone"
  RESNETS:
    OUT_FEATURES: [ "res3", "res4", "res5" ]
    DEPTH: 18
    RES2_OUT_CHANNELS: 64
  SEM_SEG_HEAD:
    NAME: "MGNetSemSegHead"
    IN_FEATURES: ["res3", "res4", "res5"]
    COMMON_STRIDE: 8
    HEAD_CHANNELS: 256
    NUM_CLASSES: 20
    LOSS_TYPE: "ohem"
    OHEM_THRESHOLD: 0.7
    OHEM_N_MIN: 262143  # (crop_width x crop_height / 4) - 1, i.e. (1024 x 1024 / 4) -1
  INS_EMBED_HEAD:
    HEAD_CHANNELS: 256
    CENTER_LOSS_WEIGHT: 200.0
    OFFSET_LOSS_WEIGHT: 0.01
DATASETS:
  TRAIN: ("cityscapes_fine_scene_seg_train",)
  TEST: ("cityscapes_fine_scene_seg_val",)
SOLVER:
  AMP:
    ENABLED: True
  WEIGHT_DECAY: 0.0
  WEIGHT_DECAY_NORM: 0.0
  WEIGHT_DECAY_BIAS: 0.0
  BASE_LR: 0.0001
  MAX_ITER: 60000
  IMS_PER_BATCH: 12
  CLIP_GRADIENTS:
    ENABLED: True
    CLIP_TYPE: "full_model"
    CLIP_VALUE: 0.01
    NORM_TYPE: 2.0
INPUT:
  FORMAT: "RGB"
  MIN_SIZE_TRAIN: (512, 640, 704, 832, 896, 1024, 1152, 1216, 1344, 1408, 1536, 1664, 1728, 1856, 1920, 2048)
  MIN_SIZE_TRAIN_SAMPLING: "choice"
  MIN_SIZE_TEST: 1024
  MAX_SIZE_TRAIN: 4096
  MAX_SIZE_TEST: 2048
  CROP:
    ENABLED: True
    TYPE: "absolute"
    SIZE: (1024, 1024)
    RANDOM_PAD_TO_CROP_SIZE: True
  IGNORED_CATEGORIES_IN_DEPTH: ["ego vehicle", "sky"]
DATALOADER:
  NUM_WORKERS: 10
TEST:
  EVAL_PERIOD: 5000
VERSION: 2
CUDNN_BENCHMARK: True