iBVP_PURE_iBVPNet_BASIC.yaml

BASE: ['']
TOOLBOX_MODE: "only_test"  # "train_and_test"  or "only_test"
TEST:
  METRICS: ['MAE', 'RMSE', 'MAPE', 'Pearson', 'SNR', 'BA', 'MACC']
  USE_LAST_EPOCH: False
  DATA:
    FS: 30
    DATASET: PURE
    DO_PREPROCESS: False               # if first time, should be true
    DATA_FORMAT: NCDHW
    DATA_PATH: "/mnt/sda/data/jitesh/PURE_Dataset/raw"                     # Raw dataset path, need to be updated
    CACHED_PATH: "/home/jitesh/data/PURE_Dataset"    # Processed dataset save path, need to be updated
    EXP_DATA_NAME: "PURE_Raw_160_72x72"
    BEGIN: 0.0
    END: 1.0
    PREPROCESS:
      DATA_TYPE: ['Raw']         #if use iBVPNet, should be Raw
      LABEL_TYPE: Standardized
      DO_CHUNK: True
      CHUNK_LENGTH: 160
      CROP_FACE:
        DO_CROP_FACE: True
        BACKEND: 'HC'    # HC for Haar Cascade, RF for RetinaFace
        USE_LARGE_FACE_BOX: True
        LARGE_BOX_COEF: 1.5
        DETECTION:
          DO_DYNAMIC_DETECTION: True
          DYNAMIC_DETECTION_FREQUENCY : 30
          USE_MEDIAN_FACE_BOX: False    # This should be used ONLY if dynamic detection is used
      RESIZE:
        H: 72
        W: 72
DEVICE: cuda:0
NUM_OF_GPU_TRAIN: 1
LOG:
  PATH: runs/exp
MODEL:
  DROP_RATE: 0.1
  NAME: iBVPNet
  iBVPNet:
    CHANNELS: 3
    FRAME_NUM: 160
INFERENCE:
  BATCH_SIZE: 4
  EVALUATION_METHOD: "FFT"        # "FFT" or "peak detection"
  EVALUATION_WINDOW:
    USE_SMALLER_WINDOW: False        # Change this if you'd like an evaluation window smaller than the test video length
    WINDOW_SIZE: 30        # In seconds
  MODEL_PATH: "./final_model_release/iBVP_iBVPNet.pth"