MitoEM-R-Base.yaml

SYSTEM:
  NUM_GPUS: 1
  NUM_CPUS: 1
MODEL:
  ARCHITECTURE: unet_plus_3d
  BLOCK_TYPE: residual_se
  INPUT_SIZE: [17, 225, 225]
  OUTPUT_SIZE: [17, 225, 225]
  IN_PLANES: 1
  NORM_MODE: sync_bn
  FILTERS: [32, 64, 96, 128, 160]
DATASET:
  IMAGE_NAME: ["im_train.json"]
  LABEL_NAME: ["mito_train.json"]
  INPUT_PATH: datasets/MitoEM_R/ # or your own dataset path
  OUTPUT_PATH: outputs/MitoEM_R/
  PAD_SIZE: [4, 64, 64]
  DO_CHUNK_TITLE: 0
  DATA_CHUNK_NUM: [4, 8, 8]
  DATA_CHUNK_ITER: 10000
SOLVER:
  LR_SCHEDULER_NAME: WarmupCosineLR
  BASE_LR: 0.04
  ITERATION_STEP: 1
  ITERATION_SAVE: 5000
  ITERATION_TOTAL: 150000
  SAMPLES_PER_BATCH: 2
INFERENCE:
  INPUT_SIZE: [17, 257, 257]
  OUTPUT_SIZE: [17, 257, 257]
  IMAGE_NAME: /n/holylfs05/LABS/pfister_lab/Lab/coxfs01/pfister_lab2/Lab/donglai/eng/db/eva/2000_73728-310272.h5
  OUTPUT_PATH: outputs/MitoEM_R/test/
  OUTPUT_NAME: result # will automatically save to HDF5
  PAD_SIZE: [4, 64, 64]
  AUG_MODE: mean
  AUG_NUM: 4
  STRIDE: [8, 128, 128]
  SAMPLES_PER_BATCH: 8