Mask-Adapter / configs /mixed-mask-training /maftp /Base-COCO-PanopticSegmentation.yaml
wondervictor's picture
Upload 186 files
ba4c371 verified
MODEL:
BACKBONE:
FREEZE_AT: 0
NAME: "CLIP"
# WEIGHTS: "detectron2://ImageNetPretrained/torchvision/R-50.pkl"
PIXEL_MEAN: [122.7709383, 116.7460125, 104.09373615]
PIXEL_STD: [68.5005327, 66.6321579, 70.32316305]
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
DATASETS:
TRAIN: ("coco_2017_train_panoptic",)
TEST: ("coco_2017_val_panoptic_with_sem_seg",) # to evaluate instance and semantic performance as well
SOLVER:
IMS_PER_BATCH: 8
BASE_LR: 0.0001
BIAS_LR_FACTOR: 1.0
CHECKPOINT_PERIOD: 50000000
MAX_ITER: 55000
LR_SCHEDULER_NAME: WarmupPolyLR
MOMENTUM: 0.9
NESTEROV: false
OPTIMIZER: ADAMW
POLY_LR_CONSTANT_ENDING: 0.0
POLY_LR_POWER: 0.9
REFERENCE_WORLD_SIZE: 0
WARMUP_FACTOR: 1.0
WARMUP_ITERS: 10
WARMUP_METHOD: linear
WEIGHT_DECAY: 2.0e-05
#WEIGHT_DECAY: 0.05
WEIGHT_DECAY_BIAS: null
WEIGHT_DECAY_EMBED: 0.0
WEIGHT_DECAY_NORM: 0.0
STEPS: (327778, 355092)
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: 1024
MIN_SCALE: 0.1
MAX_SCALE: 2.0
MIN_SIZE_TEST: 896
MAX_SIZE_TEST: 896
FORMAT: "RGB"
DATASET_MAPPER_NAME: "coco_panoptic_lsj"
TEST:
EVAL_PERIOD: 5000
DATALOADER:
FILTER_EMPTY_ANNOTATIONS: True
NUM_WORKERS: 8
VERSION: 2